Whether and how: Effects of international joint ventures on local innovation in an emerging economy

Research Policy 38 (2009) 1489–1503

Contents lists available at ScienceDirect

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

Whether and how: Effects of international joint ventures on local innovation in an emerging economy Ishtiaq P. Mahmood a,∗ , Weiting Zheng b,1 a b

Department of Business Policy, National University of Singapore, Singapore 117592, Singapore Department of Management and Marketing, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

a r t i c l e

i n f o

Article history: Received 14 November 2007 Received in revised form 30 June 2009 Accepted 7 July 2009 Available online 25 August 2009 Keywords: International joint venture Patenting output Business group Network density Institutional environment

a b s t r a c t Business groups in emerging economies frequently use international joint ventures (IJV) as a channel for knowledge acquisition and technology advancement. While IJVs provide a business group with access to new technology, how successful a group is in exploiting that new knowledge for innovative purposes depends on the groups’ ability to recombine new knowledge with its existing pool of knowledge and resources. The more resources a group spends in forming IJVs with foreign partners, the less resources the group has in developing and sustaining organizational mechanisms that facilitate integration of existing ideas and resources. Following this theoretical duality, we view the IJV–innovation relationship not as an “either–or” question, but as a question of whether and how. Specifically, viewing business groups as networks of loosely coupled firms, we examine how intra-group network structure and evolving institutional environment moderates the IJV–patenting relationship in Taiwan between 1981 and 1998. © 2009 Elsevier B.V. All rights reserved.

1. Introduction Technological linkages with foreign firms offer an important channel by which enterprises in developing economies can acquire technological capabilities and catch up with those in more advanced countries (Lall, 2002). However, such linkages can also turn collaborators into codependents (Miles and Snow, 1992; Singh and Mitchell, 1996), reducing emerging economy firms’ incentives to innovate as these firms may find it easier to simply rely on foreign partners for new technology rather than developing their own proprietary technology base. Despite a general consensus that international ties such as licensing, joint ventures, and acquisitions are useful as means of technology transfer (Hobday, 1995), it is less clear how effective these foreign linkages are as catalysts for local innovation (Bell and Pavitt, 1993). Specifically, we are not sure whether and how such technological linkages facilitate or constrain innovation in emerging economies. In explaining the microeconomic context for technological development in emerging economies, some prior studies have emphasized the social networks of the returning diasporas (Saxenian, 1999) while others highlighted the importance of strong state and policy incentives (Wade, 1990). With few notable excep-

∗ Corresponding author. Tel.: +65 6874 6387; fax: +65 6779 5059. E-mail addresses: [email protected] (I.P. Mahmood), [email protected] (W. Zheng). 1 Tel.: +852 3400 3921; fax: +852 2765 0611. 0048-7333/$ – see front matter © 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.respol.2009.07.003

tions (such as, Hobday, 1995; Mahmood and Mitchell, 2004; Chang et al., 2006a,b), however, scarce attention has been paid to the organizational actors which actually implemented the linkages and translated the linkages into innovation. Specifically, little empirical research with large sample data has been conducted on the role of business groups, the dominant organizational form in many latecomer economies (Amsden and Hikino, 1994; Khanna and Palepu, 1997). The widespread presence of business groups in emerging economies suggests that groups are likely to play an important role in the process of technology accumulation. While the extant literature recognizes that groups act as a conduit for technology transfer in many emerging economies (Amsden and Hikino, 1994; Kock and Guillén, 2001), it provides very little systematic evidence on the interface between international linkages and group innovativeness. In this paper we address this lacuna by examining the role of business groups as a mediator of foreign technology and local innovation. In particular, we investigate whether and how international joint ventures (IJV) affect the innovative performance of business groups. Hobday (1995) noted that as emerging economy enterprises move up the technology ladder, joint ventures with foreign multinationals become increasingly important as a channel for accessing specialized knowledge. While equity joint ventures are more effective as conduits for the transfer of complex knowledge than contract-based alliances (Kogut, 1988; Mowery et al., 1996), we recognize that not all joint ventures are intended to facilitate knowledge flows (e.g. Gomes-Casseres et al., 2006). For example, many joint ventures in emerging economies may be motivated by distribution or cheaper manufacturing costs. Consequently, we

1490

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

focus on the international joint ventures specifically intended for promoting technological learning and facilitating access to new knowledge. We examine how IJVs affect aggregate group innovativeness within 200 unique business groups in Taiwan between 1981 and 1998. Taiwan is a model for successful catch-up; it has transformed itself from being an exporter of labor intensive products in the 1970s to one of the most innovative countries in the world, ranking fourth globally in terms of number of US patents in 2003, surpassed only by the US, Japan and Germany. Taiwan is also a typical example of emerging markets where business groups play key roles (Hamilton and Biggart, 1988; Hamilton and Kao, 1990). The importance of international linkages in driving this transition (Gold, 1986) coupled with highly varied innovativeness of the groups makes Taiwan particularly interesting. Contrary to the received wisdom, we find a negative effect of IJV on groups’ innovative outputs where innovation is measured by domestic Taiwan patents. The ‘innovative costs’ of IJV, however, vary across groups with different internal structure and external institutional context. Specifically, the negative effect of IJV declines as affiliates within a group become more and more connected to each other through inter-firm ties (such as, director interlocks and equity ties) and/or as market-based institutions becomes more widespread over time. Our results are robust to both parametric and non-parametric estimation procedures. 2. Business groups, international linkages and local innovation Business groups are a common type of multi-business firm in developing economies, frequently dominating a substantial fraction of a country’s productive assets and strongly influencing technological development in their countries. Although groups vary across countries, the common definition is that business groups are sets of legally independent companies with activities in multiple industries, that are linked as affiliates through persistent informal links and formal relationships such as equity, director, and buyer–supplier ties (Hamilton and Biggart, 1988; Khanna and Rivkin, 2001). Like conglomerates, a group provides a corporate financial structure that controls autonomous businesses in multiple industries (Williamson, 1985). On the other hand, groups are similar to multidivisional form, in which businesses within a corporation operate with greater interdependence (Chandler, 1997). Yet groups also differ from conglomerate and multidivisional corporations. Groups are more stable and coordinated than conglomerates, while being less centralized than their multidivisional counterparts (Granovetter, 1995). Member firms in groups have separate legal status and governance systems. While they coordinate business with each other, they also are responsible to their own governance bodies such as shareholders, directors, and auditors. Thus, groups are neither pure conglomerates nor pure multidivisional firms. Instead, groups in emerging economies are more like networks of loosely coupled firms linked by formal and informal ties (Granovetter, 1995). In this study, we focus on three types of formal intra-group ties: buyer–supplier ties, director ties, and investment ties. Buyer–supplier ties arise when affiliates engage in buyer–supplier relations. Director ties arise when an individual sits on the board of multiple affiliates. The third type of ties is investment ties, which arise when affiliates own equity stakes in each other through cross-shareholding. To the extent that groups act as networks, we argue that the innovative benefits of IJVs will depend on the structural patterns of ties that bind various affiliates within a group to each other; specifically, on the density of network ties. Central to our arguments is the recognition that innovation requires two things: an access to

new ideas as well as well as an ability to integrate those ideas with existing resources and ideas (Eisenhardt and Martin, 2000). While external linkages offer a way to access ideas from outside a group’s boundary, a group’s ability to integrate those ideas is shaped by the level of intra-group network density. To the extent that innovation requires both new ideas as well as integration of ideas and resources, we hypothesize that the most innovative groups are those where external ties coincide with a relatively dense network of intra-group ties. As a second set of contingency, we examine how the innovative impacts of IJVs might vary across different institutional contexts. To the extent that foreign firms’ willingness to share new ideas as well as local groups’ incentives to develop their own technology base is likely to depend on the quality of intellectual property rights, a group’s ability benefit from IJVs will be shaped by level of institutional development of the local economy. While the proponents of the national innovation systems (Lundvall, 1998) observe that the form of organization ideal for conducting innovation vary across countries and over time, relatively little attention has been paid to understanding how institutional infrastructures moderate the effects of external ties on innovation in emerging economies. 2.1. Group-level innovation and IJV: the benefits and constraints Most business groups in emerging economies did not start out with a set of core proprietary technologies (Amsden and Chu, 2003). Groups such as Hyundai in Korea, Tata in India, or Acer in Taiwan had to rely on linkages including licensing, joint ventures, or alliances with foreign firms as a way to move up the technology ladder (Hobday, 1995). Recent innovative success of Korea and Taiwan, however, poses an intriguing paradox: the faster an emerging country catches up with the developed economies, the sooner it needs to make the transition from being an imitator to an innovator. The mechanisms that allow business groups to succeed as imitators might not be the same mechanisms that make them successful as innovators. While licensing is often very useful for acquiring basic designs and blueprints (Amsden, 1989), the technology transferred through such contractual agreements tend to be relatively old, making licensing less effective as a catalyst for innovation (Bell and Pavitt, 1993). As the need for innovation grows, other forms of linkages become more relevant as conduits for the transfer of complex knowledge than contract-based alliances (Kogut, 1988; Mowery et al., 1996). Hobday (1995) also noted that as enterprises from Korea and Taiwan moved up the technology ladder, instead of focusing on acquiring generic technologies through licensing, the more successful enterprises attempted to access specialized knowledge through joint ventures or outright acquisition of technologically sophisticated firms in developed economies. For instance, consider the case of Acer Group, Taiwan’s largest personal computer maker. In 1989, Acer allied with Texas Instruments (TI) and forged a joint venture based in Hsin-chu Science Park, Texas Instruments-Acer (TI-Acer). Although TI contributed only about 26% capital to the joint venture, it transferred its 4M DRAM technology into the joint venture and brought Acer’s IC technology into the mainstream and up to worldclass industry standard. TI-Acer eventually became one of the major sources of profit for Acer Group in 1995. While international joint ventures facilitate a firm’s search for new ideas originating outside the firm’s organizational boundary, IJVs can as well inhibit a group’s innovative performance. It is because maintaining IJVs incurs costs. The resources spent in maintaining external ties constrain a group’s ability to develop its own organizational routines that facilitate the integration of existing ideas and resources. For example, as the number of collaborative projects rises, firms encounter increasing management cost, duplication and knowledge-transfer costs (Ahuja, 1996). Instead of

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

developing their own innovative capabilities, groups may choose to continue relying on foreign technology. Once again, let us consider the case of the Acer group. While IJV with TI allowed Acer to upgrade its technological capability, conflicts gradually emerged between the collaborators. TI-Acer’s plans to produce IC logic chips were delayed because TI disagreed with this direction. To worsen the situation, during 1996–1998, excess global supply resulted in considerable price drop in DRAM, which in turn turned TI-Acer into a major cash drain. The resulting losses led to the dissolution of the partnership in 1998, when Acer bought out TI’s share. The termination of collaboration and following acquisition of TI-Acer diverted time and money that could otherwise have been available for Acer’s research and development activities. The core point of this example is that external linkages can create both benefits and constraints for group innovation. As the number of IJV increases, a group may have greater access to new ideas but faces more and more problems integrating those ideas as well as less and less incentive to develop its own core technology. Thus, we view the IJV–innovation relationship not as an “either–or” question, but as a question of whether and how. In what follows, we argue that the effect of external linkages will depend on groups’ internal tie structure, as well as the external institutional environment. 2.2. Group-level innovation: the role of internal tie structure Innovation needs more than just access to new ideas; innovation also needs a particular set of resources including financial and human capital. In developed economies, market institutions provide firms with access to these resources. However, weak institutional infrastructures in emerging economies imply that affiliates need to rely on internal markets within groups for these resources (Khanna and Palepu, 1997). To the extent that intra-group ties act as channels though which different resources are shared among affiliates within a group, a group’s ability to turn new ideas into innovations is likely to depend on the patterns of intra-group ties, specifically on the structural pattern of ties. Network literature is ambivalent about which type of networklevel tie-structure most suitable for the performance of the network as a whole. The debate is often cast in terms of the relative attributes of open versus closed networks. In line with the network literature which often uses density as a way to capture the idea of network closure (Zaheer and Bell, 2005), we consider the impact of intra-group network density on group innovation. We measure intra-group density by calculating the proportion of a group’s affiliated that have direct connections (Freeman, 1977). Network literature identifies at least three ways in which a highly dense network structure can facilitate group innovation. First, high density of inter-firm ties within a group allows the group to better integrate knowledge located in its different member firms and leverage the knowledge for the innovative capability of the group as a whole (Hansen, 1999). Second, high density facilitates the sharing of financial, technological and human resources for innovation, making it easier for a group to combine ideas and resources. Third, high network density helps create a shared understanding that makes it less risky for people within units of the network to trust one another (Granovetter, 1985). Trust encourages innovation by reducing the need for rigid control system. Freedom from rigid rules and job definitions inspires creative thinking and novel ways of knowledge combination, which is critical for generating new ideas (Schollhamer, 1982). However, as the group network density increases, firms within a group face the limits of local search as they tend to seek new ideas within an organization’s existing activities, rather than looking more broadly (Helfat, 1994; Stuart and Podolny, 1996). In the absence of new ideas and resources from outside the group, the information and resources available through intra-group ties tend

1491

to be homogenous and redundant (Burt, 1992), constraining a group’s ability to innovate. As suggested by Gargiulo and Benassi (2000), a dense network results in firms within a network to be “trapped in their own net” (p. 183). Consequently, as the density of intra-group ties increases, a group’s ability to integrate group-wide pool of ideas and resources increases while its access to new ideas declines. As the preceding discussion highlights, the effects of intra-group network density on group innovativeness can go either way. Potential complementarities between external linkages and internal ties warrant our focus on the interactions between the two, examining how the network structure of ties within groups shapes the effect of external linkages on aggregate group innovativeness. 2.3. Intra-group network structure shapes the effects of IJV on group innovativeness High group network density can increase the innovation related benefits of IJVs either by offsetting some of the constraints imposed by IJVs on group innovativeness and/or by enhancing some of the benefits of IJVs towards group innovativeness. Recall that as the number of IJV increases, it becomes increasingly difficult for a group to integrate the ideas and resources across multiple affiliates. High group network density reduces the cost of integration by allowing groups to develop “routines” or “procedures” as organizational assets. To the extent that these assets reduce the cost of transferring ideas and resources across affiliates, high density compensates for some of the negative effects of IJVs. For instance, through repeatedly borrowing new technology and entering different fields, groups in emerging economies are sometimes able to internalize fungible skills in terms of ‘project-execution capability’ which acting as a shared resource across affiliates reduces the costs of subsequent entry by groups into lines of new businesses (Amsden and Hikino, 1994). At the same time, by providing greater access to information spillovers and complementary resources within a group’s boundaries, high group network density increases a group’s expected payoff from investing in R&D. A group where affiliates are bound with each other in a dense network of investment ties is in a position to reallocate cash from businesses that have positive cash flows to new ventures with negative cash flows. Similarly, access to buyer–supplier ties increase the payoff to uncertain R&D by increasing the probability that new products and processes from corporate R&D can be commercialized inside the firm. More recently, scholars such as Kodama (1986) and Teece (1986) have suggested that diversified firms can more readily develop and commercialize ‘fusion’ technologies that involve the melding of technological capacities relevant to disparate lines of business. However, such fusion does not happen automatically. It needs an organization structure which encourages consistent communication and sharing of resources. To the extent that a highly dense network provides rich opportunities for sharing ideas and resources across affiliates, high density of intra-group ties increases a group’s incentives for developing its proprietary technology rather than continuing relying on foreign partner for new technology. Thus, high density of intra-group ties increases the net benefit from IJVs by relaxing some of the constraints of IJVs that we mentioned earlier. Similarly, to the extent that IJVs provide groups with access to ideas beyond their boundaries, some of the negative effects of local search can be offset. Because the combination of density and IJV address the dual constraints of integration cost and local search, the net innovative benefits of IJV will increase with density, where the net innovative benefits are the degree to which a given number of IJVs contributes to business group innovativeness. Therefore:

1492

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

Hypothesis 1 ((contingency effects of intra-group network structure):). The greater the intra-group network density within a group, the higher the net innovative benefits of international joint ventures. The complementarities between group density and IJVs, therefore, imply that the combination of intra-group ties and external international joint ventures most suitable for group innovation is the one where IJVs coincide with a dense network of ties. The other IJV–density combinations offer weaker conditions for the innovation. For instance, the combination of small number of IJVs and high network density offers integration opportunities but little benefits in terms of access to new ideas. Similarly, the combination of large number of IJVs and low network density offers high benefits in terms of new ideas, but few integration opportunities. 2.4. Groups’ external institutional context shapes the effects of IJV on group innovativeness The institutional environment in which a group operates is the second contingency that will affect the IJV–innovation relation. Emerging economies suffer from varying degrees of institutional under-development and the lack of institutions necessary for markets to function smoothly (Aoki, 2001; Alston et al., 1996), such as underdeveloped financial markets, financial intermediaries, scarce management talent, unreliable property rights protection and inefficient judicial systems (Khanna and Palepu, 1997; Kock and Guillén, 2001). These institutional differences also partially explain why and how technological advancements might vary across nations (Lundvall, 1998; Nelson, 1993; Nelson and Nelson, 2002) as well as across firms (Casper and Matraves, 2003). Examples of institutions particularly relevant for innovation are the rules and systems that guide the availability and quality of human, social and venture capital that provide the content and infrastructure for generating and sharing information and financial resources; legal system that protect property rights (Cooke, 2001; Edquist, 1997; Furman et al., 2002). Inadequate development of these market institutions restricts the availability of resources for conducting R&D and limits its appropriability, thus discouraging firms from investing in R&D. We believe that variation in institutional development affects how domestic firms benefit from their foreign partners. Specifically, strong institutions increase the net benefits of IJVs with regards to innovation in two ways: by increasing foreign partners’ willingness to transfer good ideas, and/or by strengthening local partners’ incentives to innovate. For example, weak intellectual property rights (IPR), poor enforcement of contracts and an uncertain legal framework make foreign firms less willing to transfer sensitive technology to their local partners, and even when they do, the quality of the technology they provide may not be of very high quality. To the extent that strong IPR encourages foreign partners to share good ideas, strong IPR increase local partners’ innovative opportunities. Specific institutional developments that protect intellectual property and facilitate the access to innovation infrastructures also increase local firms’ incentives to invest in R&D. Thus, net innovative benefits of IJVs are likely to increase as market institutions become more developed. Hypothesis 2 ((contingency effects of external institutional context):). The greater the level of institutional development, the higher the net innovative benefits of international joint ventures. Together, Hypotheses 1 and 2 predict how the effects of international joint ventures (IJV) on business groups’ innovative performance are shaped by groups’ internal network structures and external institutional environments. Two extensions of above arguments seem logical. First, it is possible that the moderating effects

Fig. 1. Summary of hypothesized relationships (solid lines indicate main relationship, dotted lines indicate moderating relationship).

of group density and institutional context will vary depending on the type of tie. In the absence of strong theoretical priors, we view this as an empirical issue. Second, as market institutions become more developed, affiliates can rely on external market transactions to access various resources (Chang et al., 2006a,b). As the access to internal markets becomes less critical as a source of innovation infrastructures, the benefits of group density as a conduit for groupwide pool of resources is likely to decline, suggesting that the role of density as a moderator will become weaker as institutions become more developed. Thus, Hypothesis 3 ((2nd order contingency effects):). The greater the level of institutional development, the weaker the moderating effects of network density. By specifying the underlying theoretical mechanism by which density (Hypothesis 1) and institutions (Hypothesis 2) jointly moderate the effects of IJV, Hypothesis 3 provides a way to tease out the robustness of our overarching logic. The hypothesized relationships are summarized by Fig. 1. 3. Taiwan: empirical setting The rise of Taiwan from an exporter of labor intensive products to becoming an innovative powerhouse warrants a greater understanding of the organizational drivers behind this transition, especially the role played by Taiwanese business groups. As Table 1a shows, the number of total domestic patent applications by affiliates of business groups rose substantially during the our analysis period, from about 33 domestic Taiwanese patents in 1981 to about 1141 in 1998. While SMEs and government financed institutes such as the Industrial Technology Research Institute (ITRI) remain important players in Taiwan in terms of their overall patent share (Mahmood and Singh, 2003), business groups have emerged as important innovators. Specifically, business group affiliates accounted for approximately 40% of Taiwan based recipients of US patents between 1990 and 1999, while seven of the top ten most innovative firms in Taiwan in terms of US patents over 1970–1999 were affiliates of business groups such as, UMC, TSMC, Walsin Lihua, Hon Hai, Mosel Pacific, and Acer. Appendix Table A1 provides a list of the top 25 most innovative groups in terms of successful domestic patent application in Taiwan in 2005. Many of the top innovators are in sectors which include semiconductor (TSMC), electronics (Hon Hai), and computing (Acer); in addition, some of the groups are leading patentees in sectors such mobile communications (Inventec), electrical machinery (Teco), chemicals (Shinkong). There is substantial variation in patenting for different groups within similar set of industries (for example, TSMC vs. UMC in semiconductors; Hon Hai vs. Kinpo in

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

1493

Table 1a Distribution of patents of Taiwan Business Groups, 1981–1998 (N = 456). Year

# of Business groups

# of manufacturing affiliates

Total domestic patent

New invention patents

Process patents

1981 1986 1990 1994 1998

84 83 81 84 124

579 634 668 774 1326

33 111 252 369 1141

2 15 65 179 763

31 96 187 190 378

Total

456

3981

4122

2243

1879

Table 1b Distribution of international linkages, 1981–1998 (N = 456). Non-learning IJV

Leaning IJV

Total IJV

No.

Share

Share

No.

Licensing No.

Share

Acquisition

Other linkages

Total

Share

No.

Share

Share

No.

1981 1986 1990 1994 1998

18 16 29 28 25

25.35% 12.50% 11.46% 14.00% 28.09%

19 24 67 49 29

26.76% 18.75% 26.48% 24.50% 32.58%

37 40 96 77 54

52.11% 31.25% 37.94% 38.50% 60.67%

29 60 98 66 19

40.85% 46.88% 38.74% 33.00% 21.35%

4 13 14 19 10

5.63% 10.16% 5.53% 9.50% 11.24%

1 15 45 38 6

1.41% 11.72% 17.79% 19.00% 6.74%

71 128 253 200 89

Total

116

15.65%

188

25.37%

304

41.03%

272

36.71%

60

8.10%

105

14.17%

741

electronics; Acer vs. Asus in Computing). The substantial variation in terms of patenting over time as well as across business groups warrants an understanding of the sources of such variations. 3.1. External technological linkages and intra-group network structure Access to international technological linkages provides a potential explanation as to why some groups might have been more innovative than others. Table 1b indicates that the absolute number of IJVs formed by groups reached its peak in early 1990s. As groups have moved up the technology ladder, it appears that the proportion of IJVs as a share of total international linkages has risen over time while the share of licensing agreement has fallen. This is consistent with some of the descriptive studies which suggest that Taiwan has already moved from OEM to ODM. There is also much variation in the number of IJVs business groups are involved in. While some business groups, such as China Rebar Group, have no international joint venture at all, whereas others, such as Far Eastern Group, can have as many as 10 IJVs. While some of the most innovative groups (UMC) relied heavily on IJVs,2 not all Taiwanese groups which have a high number of IJVs are innovative.3 Perhaps an important missing piece to the IJV–innovation puzzle lies in considering how inter-firm ties among various affiliates within a business group might contribute to or constrain innovation. Governance of groups in Taiwan involves substantial variety in the structure of intra-group ties. First, affiliates of Taiwanese groups commonly engage in buyer–supplier relationships with each other in order to take advantages of group-wide internal markets. Second, Taiwanese groups usually set up chains of equity shareholding

2 The UMC Group for instance established a joint venture with DuPont Photomasks in 1998 with the objective of producing photomasks in Taiwan. This venture, according to H.J. Wu, president of UMC, was to provide UMC Group with access to advanced binary, optical proximity correction and phase shift masks necessary to help accelerate their innovative capability. Another example is the Far Eastern Group, which set up a joint venture with AT&T in 1998; Far EasTone, which provided technology on fully integrated GSM dual-band network, facilitated Far Eastern’s successful expansion into the high-tech telecommunication field. 3 For instance, the Sampo Group, a major player in electronics, have documented 12 joint ventures with multinational companies during 1980s–1990s, yet only achieved 26 patents in 1998, far less than other innovative groups with lower level of IJVs.

No.

No.

ties among their member firms (La Porta et al., 1999; Claessens et al., 2000), which allows information access and control over selecting key personnel such as boards of directors and CEOs in affiliate firms. Third, Taiwanese groups typically hire professional managers to oversee routine administration of affiliates (Chung, 2001), while exercising strategic control through interlocking directorates of family members who often hold the position of board chairs of the affiliates. Unlike Japan where group boundaries tend to be ambiguous (Saxonhouse, 1993; Weinstein and Yafeh, 1995), in Taiwan strong cultural foundations such as patrilineal family connections and regional kinship delineate group boundaries clearly (Numazaki, 1986). The clarity of group boundary along with detailed data on inter-firm ties within a group allows us to measure group network density by calculating the ratio of actual number of ties to the number of potential ties that affiliates could forge within that group. According to Table 1c, the density based on buyer–supplier ties has dropped sharply over time, while the investment density and the director density remained relatively stable. Buyer–supplier density fell as groups in the 1990s became increasingly diversified (Amsden and Chu, 2003). While groups relied on investment ties to finance their expansion, they maintained strategic control through interlocking directorates of family members who often hold the position of board chairs of the affiliates. 3.2. Institutional environment in Taiwan The economic and political deregulation in Taiwan started from the late 1980s (Luo and Chung, 2005). Since then, market-oriented regulations (e.g. the Statute for the Transfer of Public Enterprises to Private Operation) were rapidly established (Pistor and Wellons, 1998), restrictions on loans and foreign competition were removed, and strategic industries were deregulated (Luo and Chung, 2005). Taiwan’s experience in going through substantial institutional transition during our analysis period also makes Taiwan an ideal setting for testing the moderating role of institutions. As shown in Table 1c, market-supporting infrastructures have also been gradually installed. Specific to innovation infrastructures, the Taiwanese government took the lead in setting up a venture capital industry in the 1980s (Amsden and Chu, 2003) and a science park to provide smaller firms with access to various innovation infrastructures to conduct R&D. In the late 1980s, a specialized over-the-counter (OTC) stock market for high-tech

a Market development is the mean value of the multipliers of each of the four dimensions, using 1981 as the base year (the multipliers are in parentheses); the results are robust to other combinations of the four dimensions. We borrow this measure from Mahmood et al. (2006). b The 1981 and 1986 values of the legal protection score are estimates (the market development measure is robust to alternative estimates).

32,102 (1.0) 39,065 (1.2) 49,399 (1.5) 68,274 (2.1) 87,421 (2.7) 608,658 (1.0) 741,887 (1.2) 863,664 (1.4) 975,549 (1.6) 1,034,328 (1.7) 13.2 (1.0) 39.0 (3.0) 232.3 (17.6) 351.2 (26.6) 612.0 (46.4) 1 1.7 5.5 8.1 13.2 0.43 0.4 0.37 0.35 0.39 6.8 7.6 8.2 9.2 10.7 1981 1986 1990 1994 1998

0.34 0.27 0.2 0.15 0.11

0.22 0.23 0.24 0.28 0.27

External labor market: number of university graduates Commercial intermediaries: no. of for-profit organizations Capital markets: stock market trading volume ($US million) Market development of the institutional environment: relative magnitudea Mean investment tie density Mean director tie density Mean buyer–supplier tie density Mean no. of affiliates Year

Table 1c Density and institutional context, 1981–1998 (N = 456).

3.3 (1.0) 4.3 (1.3) 5.3 (1.6) 6.4 (1.9) 7.3 (2.2)

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

Legal protection: no. of points in world competitiveness reportb

1494

firms, modeled after NASDAQ, was also established (Chang et al., 2006a,b). Other innovation-supporting institutions such as public research institutes, universities, and venture capital firms were also gradually established in Taiwan during 1980s to 1990s. These changes offer an excellent opportunity to study how institutional context shapes the effects of IJV on group innovative performance. 4. Methods 4.1. Data source Our primary data source is the directory of Business Groups in Taiwan (BGT), a data source used widely by scholars (Khanna and Rivkin, 2001; Chung, 2001). The directory of Business Groups in Taiwan is compiled by China Credit Information Service in Taipei (CCIS), the oldest and most prestigious credit-checking agency in Taiwan and an affiliate of Standard & Poor’s of the United States. CCIS started publishing data for the top 100 business groups (in terms of annual sales) biennially in 1972. For credit checking in the private sector, CCIS maintains a database containing more than 30,000 largest firms in Taiwan. It constructs the database of business groups by examining the inter-organizational relationships such as shared identity, cross-shareholding and interlocking directorate among these firms. In addition to self-identification, firms have to meet the following objective criteria to be considered as member firms, including having overlaps of shareholders, directors, auditors, or decisionmakers with the core firm and having substantial proportion of shares held by other group members. BGT defines a business group as “coherent business organization including several independent firms.” Since its second edition (which was published in 1974), BGT has consistently maintained the following criteria in selection of business groups: (1) more than 51 percent of the ownership was native capital; (2) the group had three or more independent firms, (3) the group had more than NT$100 million group total sales, and (4) the core firm of the group was registered in Taiwan. This directory is the most comprehensive and reliable source for business groups in Taiwan. According to BGT, the top 100 groups contributed 42% of national GDP in the 1990s, representing material business activity within Taiwan. Several previous studies rely on this source (Hamilton and Biggart, 1988; Khanna and Rivkin, 2001), although none has translated and coded the intra-group ties data. For group-level innovation data, we collect information about domestic patenting in relevant years from 1982 to 2000 by affiliates of business groups from online databases of the Intellectual Property Office of Taiwanese government (http://www.patent.org.tw). Taiwan’s patent system was established in 1945 and is a combination of the US and Japanese patent systems. Taiwanese patent examiners follow standards similar to their US counterparts regarding what constitutes a patentable invention (Yang, 2008).4 Patenting is common across multiple industry sectors in Taiwan, and Yu’s (1998) study of Taiwanese firms located the Hsinchu Science Park finds substantial effectiveness across a wider variety of industries than in the US (e.g. US studies by Levin et al. (1987) and Cohen et al. (2000) show that patents are most effective in the chemical and pharmaceutical sectors).

4 The major difference between the US and the Taiwanese patent system before the 1994 patent reform was the length of a patent. In accordance to the Trade Related Aspects of Intellectual Property Rights (TRIPS) agreement of the World Trade Organization, Taiwan restructured its patent systems in 1994, extending a patent’s life from 15 to 20 years.

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

4.2. Sample The BGT database provides information about business groups in Taiwan for five financial years: 1981, 1986, 1990, 1994, and 1998. Because the composition of the top 100 list changes from year to year, our sample consists of 456 group–year entries (200 unique groups). Seventy-four percent of total group revenue comes from manufacturing sectors, although almost half the affiliates operate in service sectors. We included service firms in the measures of ties on the premise that service activities can contribute knowledge needed for innovation, but we included only manufacturing sector affiliates in the innovation analyses because service businesses rarely take out patents. The final sample consists of 442 group-year observations after deleting all the missing values. 4.3. Variables 4.3.1. Dependent variable We measure Total Patents, as the count number of approved patent applications, or granted patents, for group i in year t. The particular year t is assigned as the application date for each granted patent. For example, a patent filed in 1982 but was granted in 1984 is treated as a 1982 patent. This coding procedure reflects the research output by the relevant technological efforts. As innovation output, patents thus correspond to the research efforts immediately before the patent application. We also use a one-year lag of all the other explanatory variables as influences on innovation. Thus, we regress the 1981 values of covariates and controls such as number of IJVs on the patent count for 1982. Two caveats are in order. First, patent data may fail to capture the cumulative and incremental aspect of learning. Critics further argue that many technological developments in Asia are not patented, thus making patents a less reliable source for measuring relative technological competence (Amsden and Hikino, 1994). On the other hand, the absence of uniform accounting standard across firms as well as unavailability of R&D expenditure data makes the R&D data analysis less practical in the context of many emerging economies. To the extent that patents and R&D are shown to be highly correlated (Griliches, 1990), patents can be seen more as a proxy for firm’s involvement in innovative activities than actual innovative outputs (Cohen and Levin, 1989). While we use patents as our main measure for innovation, we also test the effects on R&D intensity in our sensitivity analysis. Second, we recognize that domestic patents alone might not reflect the actual level of group innovativeness. To the extent that many large groups in Taiwan are likely to have other means for deterring potential rivals from entering (such as access to deep pockets that enable them to drive out their competitors with preemptive price-cutting in focal businesses), the large groups may not need to rely on patenting as a way to erect entry barriers. This problem, however, does not arise in the case of US patents because even the largest Taiwanese groups are not large enough to erect global entry barriers. On the other hand, patenting internationally is more expensive than patenting domestically, suggesting that the use of US patents might bias our analyses in favor of the large groups with more resources or towards groups that export more heavily to the US. To summarize, while US patenting might over-represent innovations by larger groups, local patents might over-represent innovations by smaller groups. One of the advantages of using local patents, however, is that the domestic patent database in Taiwan distinguishes between new product innovations versus incremental innovations, making this database uniquely useful for examining how the effects of IJV on innovation may vary, depending on the type of innovations. Although we use local patents as our primary measure of group innovation, we check the

1495

robustness of our results using US patents. The results appear similar to those using local patents. This is unsurprising since Taiwanese groups often simultaneously apply for patents in Taiwan and the US, suggesting that both types of patents are highly correlated (0.9). 4.3.2. Independent variables 4.3.2.1. IJV. The measure of IJVs is straightforward—by the count of the number of learning IJVs formed by group affiliates and MNCs for available each year over the research period. Learning IJVs are cross-border joint ventures formed by group-affiliated firms with partners from developed countries, with the purpose of accessing advanced technology. 4.3.2.2. Intra-group network density. For each group, we use information on the number of intra-group director ties to measure business network density. Group network density is the ratio of actual ties in a business group to the maximum number of potential ties within the group, for each of the three types of ties, i.e. Buyer–supplier Density, Director Density, and Investment Density. 4.3.2.3. Institutional development. Institutional development is measured annually by an aggregate measure encompassing four dimensions, namely, stock market growth (capital market development), number of university graduates (labor market development), number of for-profit firms (availability of intermediaries), and legal protection (legal environment development). 4.3.3. Control variables A set of variables addresses group- or industry-level influences on patenting output. Group Size records the logarithm form of total group assets (total assets of all member firms within that group). ROS denotes annual group return on sales. Prior Patent Stock describes the stock of prior patent applications and controls for dynamic feedback in the model. This variable considers that there is significant path dependency in patenting activities and firms that successfully patent in one time period are more likely to patent in subsequent time periods (Nelson and Winter, 1982; Stuart and Podolny, 1996).5 Product Diversification is measured by entropy index (Palepu, 1985). Suppose a group operates in N industry segments. If Pi is the share of the ith segment in the total sales of the group, the entropy measure of total diversification is defined as follows: DT =

N 

Pi log

i=1

1 Pi

Geographic Diversification is also measured by the entropy index, such that Ri is defined as the sales share of geographic region i attributed to the global sales of firm i. GDIV =

N  i=1

Ri log

1 Ri

5 Prior Patent Stocks is calculated as the depreciated sum of a group’s own past innovation, and is defined by: Gi,t = Yi,t + (1 − ␦) Gi,t − 1. To be more specific, Gi,1981 = Yi,1981 , Gi,1986 = Yi,1986 + (1 − ␦) Yi,1981 , and so on. We thus included the past stock of a firm’s own patents by applying a 20% depreciation rate (␦) and taking a log transformation. As a robustness check, we re-ran the analyses using a lagged dependent variable to control for unobserved heterogeneity. While the results for this analysis are similar to the ones described below, we prefer the prior patent stock over the lagged dependent variable since the latter is not exogenous to the system, and therefore is not as desirable a measure as the former is.

1496

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

Table 1d Distribution of business groups in major industries (N = 456, 200 unique groups). Major industry

Number of business groups

Electronics Chemicals Machinery Transportation Traditional

56 29 8 19 88

The variable, Pre-sample Mean, examines sample groups’ histories of innovation prior to 1981, and thus controls for unobserved heterogeneity in groups’ capability to patent (Blundell et al., 1995, 2002; Cincera, 1997; Cameron and Trivedi, 1998). It is defined as Pre-sample mean =

(



1970−1980

Total Patentsit)

Prob(yit+1 = 0|Xit , t) = Prob(process 1) + Prob(yit+1 = 0|Xit , t, process 2) ∗Prob(process 2) Prob(yit+1 = k|Xit , t) = [1 − Prob(process 1)] ∗ Prob(process 2), where k = 1, 2, 3, . . . Finally, as an LR test favors a ZINB mode to a ZIP model, we employ a ZINB model with robust standard errors as our major statistic method, adding both group- and time-specific fixed effects in the estimation. We include a set of inflated variables, which we believe will influence whether a group will patent, including group size, industry, profitability, product and geographic diversification. We also test the robustness of our results by using ZIP model as well as a negative binomial model with semi-robust standard errors using GEE for panel data.

11 5. Empirical results

As to industry-level control variables, technological opportunity (Tech Opportunity) in different industries is measured here by industry averaged R&D expenditure over sales, i.e. R&D intensity (Scherer, 1982). We also include the industry-weighted average of the fivefirm concentration ratio Industry C5 as a control for the group-wide level of exposure to competitive pressures. The Tech Opportunity and Industry C5 variables address industry-specific influences on group innovation. Industry dummies (groups’ major industries) also provide controls for industry specific effects. We use sector-specific dummy variables to control for variation in technological opportunity and propensity to patent, aggregating the sectors into five major classifications: Electronics, Machinery, Chemicals, Transportation, and Traditional. Table 1b indicates the distribution of groups in these industries. Four Time Dummy Variables (using 1981 as the baseline year) are also included to control for time dependency (Table 1d).

Table 2 reports summary statistics and correlations of key variables. It shows that there is a large range in group-affiliates’ patenting during the study period. Total Patents varies from zero to 854 (the number of patent applications filed in 1998 by United Microelectronics Corp., an affiliate of the UMC group, one of Taiwan’s leading maker of computer chips). There is also a great variation in number of learning IJVs formed by business groups, ranging from 0 to 10. Groups such as Far Eastern Group had as many as 10 IJVs in 1992, whereas most of the groups had less than three over our study period. In conclusion, the summary statistics indicate that business groups in Taiwan demonstrate a reasonable variation in patenting and IJV formation across different groups, as well as a rich variety of intra-group ties for knowledge sharing and diffusion. 5.1. Main effect of IJV on group innovation

4.4. The model We use robust zero-inflated negative binomial (ZINB) models to test the hypotheses. The count nature of our dependent variable (i.e. number of patents) suggests that an appropriate statistical method would adopt either a Poisson or a negative binomial approach, a variant of the former that can account for heteroscadasticity (Hausman et al., 1984; Cameron and Trivedi, 1986). Meanwhile, further screening of our data suggests that our count dependent variable (1) is characterized with excess zeros, and (2) exhibit overdispersion problem. For instance, the number of zeros occupies over half (66%) of our observations. Amongst 200 business groups, less than half (93) had experienced at least one case of successful patent applications during our entire study time period. Therefore, a simple negative binomial model, a zero-inflated Poisson model, and a zero-inflated negative binomial model are all candidates for count data with these characteristics (Drukker, 2000). As Vuong (1989) statistics (significantly positive) prefers a zero-inflated model, we choose to adopt the zero-inflated approach (including zero-inflated Poisson (ZIP) and zero-inflated negative binomial (ZINB) regression model) to handle the preponderance of zeros in dependent variable (Mullahy, 1986; Lambert, 1992). Essentially, the zero-inflated model separates two processes where zero outcomes may be generated. In process 1, the patent outcome is always zero (some firms never patent). In process 2, the usual Poisson or negative binomial process is at work (some firms may generate zero outcome in a year but not necessarily so in another year)

Model 1 of Table 3 reports the estimates of the baseline mode. Models 2 reports the results based on the “main-effect” model, which relates the number of learning IJVs to group patenting output. Models 3–5 present estimation results for models with moderation effects of group’s network densities, i.e. buyer–supplier density, director density, and investment density. Model 6 is a full model with all independent variables and moderation terms. The coefficients on IJV show that forming international joint ventures inhibits group innovation. For a unit increase in IJV, the expected count of patents decreases by a factor of 0.38,6 holding all other variables constant. This result suggests that despite our common understanding of the innovation-enhancing effect of technology joint ventures, Taiwan business groups failed to improve their innovativeness from the joint ventures they established. This finding is consistent with the argument that forming IJV with technologically advanced MNCs may inhibit firms’ incentives for innovative activities. This interesting result leads us further to examine the boundary conditions, i.e. which types of groups drive the negative impact. The results in Table 3 also indicate that two out of three types of intra-group networks ties inhibit group patent application. Both director ties and investment ties negatively influence group innovation, consistent with the argument that dense director-interlocking and equity cross-holding may inhibit group’s innovative activities as a result of redundant information and probably controlling family’s private interests.

6

From the coefficient in the “Full Model” model, exp(−0.967) = 0.38.

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

1497

Table 2 Correlation matrix of key variables. Key variables 1. Total patent 2. IJV 3. Buyer–supplier density 4. Director density 5. Investment density 6. Institutional development 7. Group size 8. Group ROS 9. Prior patent stock 10. Product diversification 11. Geographic diversification 12. Pre-sample mean 13. Tech opportunity 14. Industry C5 Mean S.D. Minimum Maximum

1 1.00 −0.02 −0.09* −0.04 −0.02 0.16 0.23* 0.10* 0.91* −0.13* 0.02 0.01 −0.02 −0.13* 9.03 56.62 0 854

2

3

4

5

6

7

8

9

1.00 −0.17* −0.14* −0.05 −0.01 0.14* 0.05 −0.01 0.02 0.52* 0.04 −0.01 0.05

1.00 0.16* 0.12* −0.34 −0.38 −0.01 −0.06 −0.34* −0.11* −0.06 0.05 0.19

1.00 0.26* −0.03 −0.18* 0.02 −0.02 −0.06 −0.01 −0.03 0.05 0.01

1.00 0.11 −0.03 0.05 −0.01 −0.08 0.00 −0.07 −0.05 0.00

1.00 0.47 −0.00 0.10* −0.05 −0.04 −0.03 −0.14 −0.50

1.00 0.08 0.19* 0.31* 0.12* 0.27* −0.08 −0.18*

1.00 0.10* −0.02 0.08 0.02 −0.05 0.03

1.00 −0.14* 0.02 0.01 −0.01 −0.13*

0.41 0.97 0 10

0.21 0.24 0 1

0.38 0.30 0 1

0.25 0.21 0 1

3.89 2.57 0.9 7.5

8.95 1.41 5.46 13.42

4.80 11.95 −24.9 29.30

5.02 47.93 0 793

10

11

12

13

14

1.00 0.01 0.09* 0.12* 0.06

1.00 0.01 −0.03 0.10*

1.00 −0.01 0.03

1.00 −0.01

1.00

0.28 0.19 0 0.85

0.12 0.23 0 0.87

0.39 2.21 0 19.80

3.87 1.41 0 5.50

0.21 0.16 0.01 0.83

N = 456. * p < 0.05.

Table 3 Results of zero-inflated negative binomial (ZINB) regression, 1981–1998 (dependent variable: total number of patents). Variables

(1)

(2) −0.393** (0.175)

IJV Buyer–supplier density * IJV

(3) −0.426** (0.185) 0.500 (1.081)

Director density * IJV

(4)

(5)

−0.891*** (0.153)

−0.721*** (0.141)

1.245*** (0.271)

Investment density * IJV Buyer–supplier density Director density Investment density Group size ROS Prior patent stock Product diversification Geographic diversification Pre-sample mean Tech opportunity Industry C5 Electronics Chemicals Machinery Transportation Constant Log pseudo-likelihood Wald Chi-square Observations

−1.852 (1.600) −1.117* (0.603) −0.962* (0.573) 0.495*** (0.074) −0.006 (0.011) 0.005* (0.003) 0.169 (0.895) 1.665** (0.738) 0.035 (0.027) −0.019 (0.064) −0.449 (0.624) 0.669** (0.328) −1.059*** (0.356) 0.278 (0.604) 0.642* (0.357) −2.599*** (0.861) −681.853 175.81*** 442

−1.771 (1.136) −1.121** (0.490) −0.728 (0.526) 0.463*** (0.115) −0.012 (0.022) 0.004** (0.002) 0.009 (1.221) 1.935** (0.827) 0.033 (0.026) 0.003 (0.069) −0.592 (0.668) 0.776** (0.302) −0.827 (0.784) 0.474 (0.690) 0.781*** (0.291) −2.760* (1.418) −680.737 85.03*** 442

−1.839* (1.102) −1.092** (0.485) −0.716 (0.529) 0.466*** (0.112) −0.013 (0.022) 0.004** (0.002) 0.005 (1.207) 1.931** (0.820) 0.033 (0.026) −0.001 (0.067) −0.562 (0.677) 0.789*** (0.306) −0.833 (0.770) 0.512 (0.686) 0.773*** (0.294) −2.785** (1.395) −678.260 75.26*** 442

1.292*** (0.352) −1.621 (1.001) −1.310*** (0.475) −1.421*** (0.542) 0.480*** (0.075) −0.010 (0.011) 0.004* (0.002) −0.013 (0.844) 1.820** (0.748) 0.030 (0.024) 0.002 (0.064) −0.802 (0.600) 0.665** (0.297) −1.059*** (0.314) 0.310 (0.601) 0.680** (0.300) −2.554*** (0.916)

−1.440 (0.941) −1.905*** (0.522) −1.208** (0.497) 0.487*** (0.074) −0.006 (0.012) 0.004** (0.002) 0.140 (0.825) 1.754** (0.712) 0.023 (0.023) −0.012 (0.061) −0.559 (0.583) 0.592** (0.284) −1.138*** (0.332) 0.147 (0.597) 0.605** (0.302) −2.442*** (0.866) −677.246 200.27*** 442

−675.226 209.01*** 442

Time dummy variables are not presented due to space constraint. We included the following variables into our inflated models: group size, industry, profitability, product and geographic diversification. Robust standard errors in parentheses; ***p < 0.01; **p < 0.05; *p < 0.1. Nonlinear effect of IJV on patenting was also tested but not detected.

(6) −0.967*** (0.160) 0.337 (0.734) 1.101*** (0.284) 0.419 (0.320) −1.512 (0.940) −1.857*** (0.525) −1.371*** (0.523) 0.505*** (0.075) −0.007 (0.011) 0.004** (0.002) 0.069 (0.823) 1.745** (0.718) 0.024 (0.023) −0.010 (0.062) −0.594 (0.593) 0.591** (0.289) −1.134*** (0.334) 0.198 (0.593) 0.590* (0.304) −2.582*** (0.896) −672.65 209.68*** 442

1498

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

Table 4 ZINB Estimates of influences on group patenting output in institutional transition. Variables IJV

(1) −0.351** (0.168)

IJV * institutions Buyer–supplier density * IJV Buyer–supplier density * institutions Buyer–supplier density * IJV * institutions

(2) −0.869** (0.395) 0.302*** (0.084) 10.071** (4.582) −0.518** (0.214) −2.282** (1.034)

Director density * IJV

(3)

(4)

−1.459** (0.568) 0.182* (0.110)

−0.723* (0.475) 0.042 (0.126)

3.163* (1.874) 0.203 (0.205) −0.427 (0.394)

Director density * institutions Director density * IJV * institutions Investment density * IJV

0.831 (1.777) 0.108 (0.269) 0.089 (0.478) −1.665* (0.985) −1.404*** (0.466) −1.703* (1.014) −0.038 (0.111) 0.466*** (0.124) −0.012 (0.027) 0.004* (0.002) 0.248 (0.897) 1.751** (0.756) 0.034 (0.024) −0.035 (0.077) −0.710 (0.960) 0.790** (0.351) 0.446 (0.615) 0.881** (0.428) −0.896 (1.091) −2.318 (2.044)

Investment density * institutions Investment density * IJV* institutions Buyer–supplier density Director density Investment density Institutional development Group size ROS Prior patent stock Product diversification Geographic diversification Pre-sample mean Tech opportunity Industry C5 Electronics Machinery Transportation Chemicals Constant Log pseudo-likelihood Wald Chi-square Observations

−1.606 (1.003) −1.216*** (0.470) −0.789 (0.544) −0.009 (0.069) 0.459*** (0.097) −0.012 (0.025) 0.004** (0.002) 0.160 (1.404) 1.862** (0.776) 0.039 (0.025) −0.017 (0.069) −0.460 (0.910) 0.915*** (0.339) 0.590 (0.690) 0.859*** (0.330) −0.754 (0.732) −2.733* (1.409) −635.77 81.79*** 442

−3.320** (1.594) −1.135** (0.493) −1.145** (0.565) −0.083 (0.090) 0.445*** (0.074) −0.017 (0.018) 0.004** (0.002) 0.392 (1.165) 1.507** (0.757) 0.039 (0.024) −0.041 (0.062) −0.848 (0.867) 0.715* (0.371) 0.677 (0.588) 0.752* (0.403) −0.861* (0.458) −1.765 (1.142) −631.16 187.39*** 442

−1.256 (0.977) −2.941** (1.178) −1.475** (0.604) −0.084 (0.113) 0.527*** (0.080) −0.004 (0.011) 0.004* (0.002) 0.408 (0.810) 1.656** (0.766) 0.039 (0.025) −0.044 (0.061) −0.418 (0.851) 0.677** (0.317) 0.252 (0.589) 0.751** (0.365) −1.032*** (0.368) −2.184** (0.953) −629.36 207.46*** 442

−632.69 94.34*** 442

Robust standard errors in parentheses; ***p < 0.01; **p < 0.05; *p < 0.1.

Several other control variables also influence group patenting output. Larger groups have more patents, consistent with common US evidence. The positive and significant coefficients on Prior Patent Stocks reinforce that innovative activities are dynamic process and prior experience in patenting promote current innovation. The positive coefficients on Geographic Diversification support the argument that geographically diversified groups have greater opportunities to learn and acquire knowledge from multiple contexts (e.g. Chang, 1995; Teece, 1992; Chung and Alcacer, 2002). As to industry-specific effects, Electronics and Transportation have positive effects, indicating a higher level of patenting in these indus-

tries; whereas groups in Chemicals industry have significantly fewer patents. 5.2. The contingency effects of group network density and institutional context In order to test the boundary conditions of the IJV–patenting relationship, we add interaction terms to the main-effect model (Models 3–5). Our Hypothesis 1 is supported by Models 4 and 5. The positive coefficients on Director Density * IJV and Investment Density * IJV are consistent with our prediction that when the intra-group

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

1499

network density is low, forming IJV has a strong, negative effect on group innovativeness; whereas the negative effect becomes weaker (or even turns positive) when the intra-group network density is high. The significant coefficient on Investment Density * IJV, however, does not hold in the full model (Model 6). Collectively, these results indicate that with a high director network density, affiliated firms are able to share and diffuse knowledge and information, as well as search for knowledge internally to develop technological capabilities instead of solely relying on their partners for innovation. Next, in Table 4, the positive coefficient on IJV * Institutions supports Hypothesis 2 – the institutional development reinforces the positive effects of IJV and weakens the negative effects of IJV on groups’ innovativeness. We further test the three-way interaction of IJV, internal ties, and external institutional development. The negative coefficient on Buyer–supplier Density * IJV * Institutions supports Hypothesis 3, suggesting that as external institutions become more developed, ability to integrate group-wide pool of resources (through dense network of intra-group ties) becomes less valuable. Fig. 2. Moderating effect of director density on IJV–patenting link.

5.3. Endogeneity problem and selection bias The question remains whether high IJV leads to low group innovation, or high level of IJVs is a function of a group’s poor innovative capability, the source of which lies somewhere else. As LaLonde (1986) has shown, standard econometric techniques for assessing treatment effects in the presence of endogeneity problems may lead to biased estimates. We address the endogeneity problem in two ways. First, we follow the propensity score approach (Dehejia and Wahba, 1998, 1999) and construct two homogeneous comparable ‘quasi-experimenting groups’ – the treatment group (groups that increased IJV) and the control group (groups that did not); we then test whether the patenting output of these two groups are significantly different. The results provided in Appendix A2 support of our findings. Second, we follow a novel econometric approach (Rajan and Zingales, 1998) that calls for identifying a specific theoretical mechanism by which IJVs are supposed to affect innovation and documenting its working. Specifically, we argue that if IJVs are meant to benefit aggregate group innovation, the benefits are likely to be stronger when a group has access to complementary resources necessary to capitalize the new ideas. Fig. 2 relies on kernel regression to view the interaction non-parametrically.7 At lower levels of director density, a higher level of IJV has a negative impact on group innovation (the right front axis), consistent with our parametric results. On the other hand, the effect becomes positive when the director density level goes higher (the left rear axis), consistent with our hypothesis. This pattern supports the idea that the internal network density helps member firms to share and diffuse knowledge and information absorbed from external linkages. The greatest virtue of this simple test is that, it looks for evidence of a

specific mechanism by which IJV affects innovation, thus providing a stronger case for causality (Rajan and Zingales, p. 560). Sample selection bias marks another potential problem with our data that can lead us to misleading interpretations. As some group may fall out of the BGT directory in a certain year, our analyses may thus be biased if the groups that exit the sample are significantly different from those that remain. We address this problem in following three ways: First, we checked whether “exiters” are significantly different from “remainers”, comparing these two groups using a ttest. The results reject the possibility that groups that fall out of the directory are significantly different from those that remain in terms of patenting output. Next, we re-estimated our models based on a balanced panel of 1986, 1990, and 1994. This approach avoids the event of early exits, and the results are consistent with our prior findings. Finally, we employed Heckman’s (1979) two-stage sample selection estimation approach and followed Wooldridge (2002, pp. 577–585), who recognizes the conditional nature of empirical relationships and allows for more meaningful interpretation of coefficient estimates in a panel data structure. Essentially, as a first stage, we predicted a group’s likelihood of remaining in the BGT directory using two important group characteristics, i.e. group size and diversification. Larger, more diversified business groups are more likely to continue being included into the BGT directory in the next year. While these two variables may not fully explain the reason for groups to remain in the directory, they are a reasonable set of influences on group existence. Then, we used the estimates of the selection equation to create a selection variable for the main ZINB equations, based on the inverse Mills ratio. The results of the two-stage regressions confirmed a negative relation between IJV and patenting output for Taiwan business groups.

7 Kernel regression does not impose parametric assumptions on the data. Instead, it creates a three-dimensional mesh diagram that demonstrates how two explanatory variables interact with each other to influence an outcome variable (Hardle, 1990). Specifically, multivariate nonparametric regression aims to estimate the functional relation between a univariate response variable Y and a d-dimensional explanatory variable X, i.e., the conditional expectation E(Y |X) = E(Y |X1 , ..., Xd ) = m(X).he multivariate Nadaraya–Watson estimator can then be written as a generalization of the univariate case. Suppose that we have independent observations (x1 , y1 ), ..., (xn , yn ), then this estimator is defined as

5.4. Additional sensitivity checks

 n  xi1 −x1 x −xp K , ..., iph yi h1 p i=1 ˆ h (x) = n m  xi1 −x1  , xip −xp i=1

K

h1

, ...,

hp

where K (•) denotes a so-called kernel function, and h the bandwidth (Hardle, 1990).

5.4.1. Using alternative measures of innovation We performed two robustness checks on our dependent variable. First, we test the effects of IJV on R&D intensity. Second, we use US patents as our dependent variable. In both cases, the results remain materially unchanged. 5.4.2. Using alternative measures of IJV We performed several robustness checks on our key independent variable, IJV. First, we also measured our key independent variable as the percentage of learning IJV to the total number inter-

1500

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

national linkages for group i in year t. Apparently, the absolute number of IJV is a more straightforward measure for importance of IJV. However, if a group has a large absolute number but relatively small share of IJV, the cross-sectional comparison to another group with a small absolute number but large share of that linkage could be biased. Thus, we also included the proportional measure of IJV as a robustness check for our analysis, using the share of IJV to total international linkages as the independent variable. The similar coefficients on proportion of IJV are consistent with our prior results. 5.4.3. Controlling for licensing As licensing provides an alternative channel for acquiring foreign technology, a firm with a large number of licensing agreements may not resort to IJV for technological advancement. Results show that even after controlling for the effect of licensing, the negative effect of IJV on group patenting remains. In addition, we tested for a non-linear effect of number of IJV on patenting in our main equation, but did not detect any. 5.4.4. Controlling for outliers One might also question whether the negative innovation effect of IJV formation in Taiwan was driven by groups such as UMC, TSMC, or Hon Hai — groups that are highly innovative but have few joint ventures. To address this issue, we performed another analysis by excluding the most innovative five groups, but our results did not change materially. 5.4.5. Controlling for differences in the “quality” of innovations We were able to differentiate two types of innovations, highnovelty innovation and low-novelty innovation.8 It is likely that different types of innovative activities require different types of assets. Therefore, besides looking at total number of patents, we also differentiate two types of patents, new invention patent and process patent. New invention patents designate new products, materials, or manufacturing processes. Process patents represent incremental changes in design or minor modification in the shape or color of a product. The results indicate negative effects of IJV on both types of patents. 6. Discussion and conclusion This study shows that international joint ventures in Taiwan have a systematic negative impact on group’s innovativeness— measured by number of patent applications. The results also indicate the importance of groups’ internal structure in moderating the main effect, i.e. the negative innovation effect of forming IJV gets weaker for groups with high internal director network density. Moreover, our results highlight the importance of institutions: domestic firms start to benefit more from their IJVs as institutions evolve; and the value of intra-group resource integration ability would also decrease as better institutions are gradually installed. In this study, we address multiple gaps in the literature. First, our study contributes to alliances literature by helping to shed light on the duality – particularly the downside – of the IJV–innovation relation. Many researchers have attempted to caution us about the potential risks of establishing external linkages under certain conditions (e.g. Baum et al., 2000; Khanna et al., 1998; Hill, 1992;

8 In the Taiwan patent database, three types of patent are included: (1) new invention, designating a wholly new product, material, or manufacturing process; (2) new model, meaning major improvement in outlook or structure of a product or production process; and (3) new style, representing a minor modification in the shape or color of a product.

Pisano, 1990; Shan, 1990). However, so far, few empirical studies have directly examined such relationship, particularly for partners from emerging economies. By highlighting the institutional importance as well as organizational contingencies driving the IJV–innovation relation, we make some progress towards unpacking the theoretical linkages by which joint ventures influence an organization’s innovative activities. Second, our study also contributes to the theory of economic development by showing that allying with foreign firms may not always be a blessing for the long-term growth of firms in emerging economies. Although strategies including collaborative strategies of multinational firms are heavily studied, far less attention has been placed on the risks of these partnerships impose on the domestic partners in these contexts. Third, we contribute to the small but growing field of research on business groups by addressing how internal structure of groups might increase or reduce groups’ ability to benefit from international joint ventures. Despite the fact that groups as an important conduit through which a disproportionate share of international linkages take place (Khanna and Palepu, 1997; Amsden and Hikino, 1994), few papers examine the interface between joint ventures and innovation in the context of groups. Our results also have important managerial and policy implications. Managers should realize that allying with MNCs for innovative purposes may not be successful, especially when it is not in line with firms’ organizational structure. From a long-term perspective, policy makers need to understand under what conditions will foreign firms benefit or not benefit the performance of local firms, thus tailor their policies to promote the long term growth of the indigenous firms. 7. Limitations and extensions The work has several limitations that could lead to additional fruitful research. First, this study relies on one country that has only partially gone through marketization; studies involving multiple countries at different points of marketization would be useful. Second, using patents as a measure of innovative activity has its own limitations, many of which are widely acknowledged in the innovation literature that uses patents as measures of innovation. It would be useful to follow-up by using alternative measures of innovation such as new product introductions. Third, it is important to examine how the collaborative strategies of firms from emerging economies, together with their performance consequences, evolve over time. An evolutionary approach (e.g. Barnett and Burgelman, 1996) tracking the ties between business groups and foreign firms, as well as the ties between groups and local SMEs would allow us to examine the direct and indirect impacts of IJVs. Such extensions would continue the task of unpacking the black box of inter-organizational linkages and innovation. Acknowledgements We appreciate comments from participants of the strategy and international business seminars at Duke University, London Business School, and the University of Michigan. We are specifically grateful to Tom Murtha for helpful comments. We also thank Zixia Sheng and Hongjin Zhu for excellent research assistance. Ishtiaq Mahmood acknowledges support from the Faculty Research Committee at NUS Business School. Appendix A. Table A1.

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

1501

Table A1 Top 25 most innovative groups based on domestic patent applications in 2005. Rank

Group name

Year of establishment

Main products

Patents

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

TSMC HON HAI BENQ INVENTEC LITEON ACER ADVANCED SEMICONDUCTOR KINPO WALSIN LIHUA VIA TECHNOLOGIES TECO ELECTRIC & MACHINERY SHIN KONG UMC MEDIA TEK WISTRON LIEN HWA MITAC CHI MEI DELTA ELECTRONICS TATUNG ASUSTEK FORMOSA PLASTICS MACRONIX SPIL QUANTA COMPUTER WPG HOLDINGS

1987 1974 1984 1975 1989 1979 1984 1973 1966 1992 1956 1952 1980 1997 2001 1955 1960 1975 1950 1990 1954 1989 1984 1988 1981

Semiconductor Electronics Computing, electronics Computing, Mobile communication Optical drive Computing Semiconductor Electronics Wire and cable; Computing Computing Electric machinery Chemical Materials Semiconductor Semiconductor Computing, Communication Computing Petro-Chemical Electronics Electronics; Machinery Computing Plastic Products Semiconductor Electronics Computing Electronics

3981 3346 1712 1479 1091 1058 868 861 674 657 636 526 495 422 406 399 381 359 338 328 321 278 248 224 162

Appendix B. B.1. Propensity score approach Let P1 represent patenting performance for a group that increased its IJVs at t + 1, and P0 the performance of a group that did not increase its IJVs at t + 1. Let E be an IJVs indicator that equals one when the group increased IJVs at t + 1 and zero otherwise. Accordingly, E(P1 |Ei = 1) denotes the average patenting performance of groups that increased IJVs at t + 1, and E(P0 |Ei = 0) the average performance of the groups that did not increase IJVs at t + 1. The effect of interest is the difference of the effect of IJV on the patenting performance of the groups that increased and that did not increase IJVs, or put differently, the difference between the patenting performance of the group (increased IJVs) and the performances of itself would have had if it did not increase IJVs (E(P0 |Ei = 1): |T =1 = E(P1 |Ei = 1) − E(P0 |Ei = 1)

(1)

to the matching methodology of propensity score, which is based on the propensity score theorem by Rubin (1974, 1977), and was further applied into practice by Dehejia and Wahba (1998). According to Dehejia and Wahba (1998), we define the propensity score as the probability of assigned treatment conditional on a vector of iid. variables Xi . P(Xi ) ≡ Pr(Ii = 1|Xi ) = E(Ii |Xi )

(3)

According to propensity score theorem, if the treatment assignment is ignorable conditional on X, it is also ignorable conditional on the propensity score: Yi1 , Yi0 ⊥ Ei |Xi ⇒ Yi1 , Yi0 ⊥ Ei |P(Xi )

(4)

Table A2 Causality test: propensity score approach—step 1 (Logit model). Variables

DV: Probability of forming an IJV

This is the difference between the expected treatment effects on the controlled. Since E(P0 |Ei = 1) is obviously unobservable, as the group had actually increased IJVs, the feasibly computable part instead of (1) is the difference in average patenting performance between those groups that increased IJVs and those did not:

Group size

|T =1 = E(P1 |Ei = 1) − E(P0 |Ei = 0)

Electronic

0.280* (0.157) 0.008 (0.009) 0.821 (0.668) 4.649*** (0.533) −0.155 (0.402) 1.521*** (0.590) 0.283 (0.481) 0.149 (0.481) −0.104 (0.085) 0.832 (0.723) −3.720*** (1.422) 444 104.20***

(2)

The problem, then, is that (2) is necessarily a biased estimator of (1) unless E(P0 |Ei = 0) = E(P0 |Ei = 1) under random assignment. Basically, most of the cause effect identification is actually frustrated by this inherent fact of observational life that is called the fundamental problem of causal inference (FPCI), i.e.: as long as it is impossible to observe the value of E(P1 |Ei = 1) − E(P0 |Ei = 0) on the same unit, it is impossible to observe the effect of E on P (Holland, 1986). However, the implicit threaten of FPCI does not necessary force us to give up too quickly. Rubin (1974, 1977) proposed a statistical solution to solve this problem. The remarkable merit of statistical solution is that it replaces the impossible-to-observe causal effect of E on a specific unit with the possible-to-estimate average causal effect of E over a population of units. This idea led

ROS Product diversification Geographic diversification

Machinery Transportation Chemicals Tech opportunity Industry C5 Constant Observations Wald Chi-square

Note: robust standard errors in parentheses. ***p < 0.01; **p < 0.05; *p < 0.1

1502

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503

Table A3 Causality test: propensity score approach—step 2. Block

[1] [2] [3] [4] [5] [6] [7] [8] Mean

Range of block

Total patents of group 1 (increased IJV)

Min.

Max.

0.003 0.157 0.183 0.209 0.242 0.278 0.370 0.445

0.157 0.181 0.207 0.238 0.255 0.348 0.427 0.561

Total patents of group 2 (not increased IJV)

# Obs. 0.994 0.357 0.682 0.400 1.000 0.056 0.400 0.556

Difference in total patents

t-stat

0.99 0.35 −0.068 −3.886 −20.85 −2.069 −0.029 −0.016

0.70 1.18 −0.111 −1.397 −1.117 −1.498 −0.059 −0.042

−2.736

−2.12**

# Obs.

8 8 8 7 7 8 7 7

0.000 0.000 0.750 4.286 21.85 2.125 0.429 0.571

0.814

3.550

166 14 22 10 9 18 5 9

Note: The effect is computed as the mean of the IJV effects within blocks of comparable treatment groups (increased IJV in time t + 1) and control groups (did not increase IJV in time t + 1), weighted by the number of observations in the block. Put simply, we compare the average innovation output (number of total patents) between business groups that did not increase their IJV and those that increased their IJV. The blocks are defined by the quantiles (or even more) of the propensity score distribution for the groups that increased IJV. Propensity scores are the predict probabilities of increasing IJV from the logistic model. The selection procedure based on propensity score has eliminated all the observations that are incomparable between treatment and control groups.

Rubin (1974, 1977) show that in the non-experimental context, we can identify the expected treatment effect on the controlled, by assuming that the assignment to treatment or control groups is a function of observable variables. In this case, conditional on the observed variables, treatment assignment can be taken to be random. Put simply, the theorem implies that observations in the non-experimental samples but with the same propensity score have the same probability distribution of the whole vector of observable variables Xi . Therefore, even for non-experimental data, we can still attain the maximum comparability between treatment and control groups by matching on the propensity score. The unconditional effect can then be estimated as the expected treatment effects conditional on the distribution of the controlled population: |T =1 = Ep(X){E(Yi1 |p(Xi ), Ei = 1) − E(Yi0 |p(Xi ), Ei = 0)|Ei = 1}

(5)

B.2. Simple algorithm for estimating the propensity score (Dehejia and Wahba, 1998, 1999) 1. Estimating the propensity to increase external linkages using a logistic model Pr(Di = 1|Xi ) = f (Xi ), where Xi are group characteristics. Appendix Table A2 presents results of the logistic regression of group variables on propensity to form an IJV. 2. Computing the propensity scores for both treatment and control observations from the predicted values in the model of step 1. Separating treatment and control groups (increased external linkages vs. did not increase external linkages). Discarding all those groups that are not comparable in terms of propensity score. 3. Classifying (using K-means clustering methods (see Johnson and Wichern, 2002) all groups (increase and non-increase) into blocks initially categorized by the quantiles of the propensity score distribution for ‘non-increase’ groups. 4. For each within-block propensity score, doing t-test and F-test of differences in means and standard errors between the ‘increased’ and ‘non-increased’ groups within each block. Repeat step 3 dividing blocks into finer blocks and re-evaluating if a block is not well balanced. 5. Estimate the effect of IJV changes on patenting performance by taking the weighted average (by number observations) of the within-block mean differences in patenting performance between ‘increase’ and ‘non-increase’ groups. This is the average treatment effect on the controlled in the causal inference literature.

Appendix Table A3 below reports the comparison results between those groups that increased IJV and those did not under the propensity score methods. We focus on the mean of total patents. Having IJVs seems to exert negative effects (6 negative vs. 2 positive) on business groups’ patenting output and the overall difference is statistically significant at 5% (t-stat −2.121). The general findings here mean that after we constructed two quasi-experimental control and treatment groups, the increase of IJV in the treatment group will generally bring about fewer corresponding patents than the control group. This supports our prior findings. References Ahuja, G., 1996. Collaboration and innovation: a longitudinal study of interfirm linkages and firm patenting performance in the global advanced materials. Ph.D. dissertation, University of Michigan. Alston, L.J., Eggerston, T., North, D.C., 1996. Empirical Studies in Institutional Change. Cambridge University Press, New York. Amsden, A.H., 1989. Asia’s Next Giant: South Korea and Late Industrialization. Oxford University Press. Amsden, A.H., Chu, W.W., 2003. Beyond Late Development: Taiwan’s Upgrading Policies. MIT Press. Published in Chinese by Linking Press, Cambridge, MA. Amsden, A., Hikino, T., 1994. Project execution capability. Organizational know-how and conglomerate corporate growth in late industrialization. Industrial and Corporate Change 3 (1), 111–148. Aoki, M., 2001. Towards a Comparative Institutional Analysis. MIT Press, Cambridge, MA. Baum, J., Calabrese, T., Silverman, B.B., 2000. Don’t go it alone: Alliance network composition and startup’s performance in Canadian biotechnology. Strategic Management Journal 21 (3), 267–294. Barnett, W.P., Burgelman, R.A., 1996. Evolutionary perspectives on strategy. Strategic Management Journal 17, 5–19. Bell, M., Pavitt, K.L.R., 1993. Technological accumulation and industrial growth: contrasts between developed and developing countries. Industrial and Corporate Change 2, 157–210. Blundell, R., Griffith, R., Van Reenen, J., 1995. Dynamic count data models of technological innovation. Economic Journal 105 (March), 333–344. Blundell, R., Griffith, R., Windmeijer, F., 2002. Individual effects and dynamics in count data models. Journal of Econometrics 108 (1), 113–131. Burt, R.S., 1992. Structural Holes. Harvard University Press, Cambridge, Mass. Cameron, A., Trivedi, P.K., 1986. Econometric models based on count data: comparisons and applications of some estimators and tests. Journal of Applied Econometrics 1, 29–54. Cameron, A., Trivedi, P., 1998. Regression Analysis with Count Data. Cambridge University Press, Cambridge, UK. Casper, S., Matraves, C., 2003. Institutional frameworks and innovation in the German and UK pharmaceutical industry. Research Policy 32 (10). Chandler, A.D., 1997. Strategy and structure. In: Nicolai, J., Foss (Eds.), Resources, Firms, and Strategies: A Reader in the Resource-based Perspective. Oxford University Press, New York. Chang, S.J., 1995. International expansion strategy of Japanese firms: capability building through sequential entry. Academy of Management Journal 38, 383–407. Chang, S., Chung, C.N., Mahmood, I.P., 2006a. When and how does business group affiliation promote firm innovation: a tale of two emerging economies. Organization Science.

I.P. Mahmood, W. Zheng / Research Policy 38 (2009) 1489–1503 Chang, S.J., Chung, C.N., Mahmood, I.P., 2006b. When & how does group affiliation promote firm innovation: a tale of two emerging economies. Organization Science 17 (5), 637–656. Chung, C.N., 2001. Markets, culture and institutions: the emergence of large business groups in Taiwan, 1950s–1970s. Journal of Management Studies 38, 719–745. Chung, W., Alcacer, J., 2002. Heterogenous investment motives and location choice of foreign direct investment in the United States. Management Science 48, 1534–1554. Cincera, M., 1997. Patents, R&D and technological spillovers at the firm level: some evidence from econometric count models from panel data. Journal of Applied Econometrics 12 (3), 265–280. Claessens, S., Djankov, S., Lang, L., 2000. The separation of ownership and control in East Asian corporations. Journal of Financial Economics 58 (1–2), 81–112. Cohen, W.M., Levin, R., 1989. Empirical studies of innovation and market structure. In: Smalensee, R., Willig, R. (Eds.), Handbook of Industrial Organization, II. Elsevier Science Publications. Cohen, W.M., Nelson W., Walsh, J., 2000. Protecting their intellectual assets: appropriability conditions and why U.S. manufacturing firms patent (or Not), NBER Working Paper No. W7552. Cooke, P., 2001. Regional innovation systems, clusters and the knowledge economy. Industrial and Corporate Change 10 (4), 945–974. Dehejia, R.H., Wahba, S., 1998. Propensity score matching methods for nonexperimental causal studies, National Bureau Of Economic Research, Working Paper No. 6829. Dehejia, R.H., Wahba, S., 1999. Causal Effects in Non-Experimental Studies: Reevaluating the Evaluation of Training Programs. Journal of the American Statistical Association 94, 1053–1062. Drukker, D.M., 2000. STATA Resources and Support, FAQs. STATACorp. Edquist, C., 1997. Systems of Innovation. Technologies, Institutions and Organizations. Pinter, London. Eisenhardt, K.M., Martin, J.A., 2000. Dynamic capabilities: what are they? Strategic Management Journal 21 (10–11), 1105–1121. Freeman, L.C., 1977. A set of measures of centrality based on betweenness. Sociometry 40, 35–41. Furman, J.L., Porter, M.E., Stem, S., 2002. The determinants of national innovative capacity. Research Policy 31, 899–933. Gargiulo, M., Benassi, M., 2000. Trapped in your own net? Network cohesion, structural holes the adaptations of social capital. Organization Science 11, 183–196. Gold, T.B., 1986. State and Society in the Taiwan Miracle, Armonk. ME Sharpe, Inc., New York. Gomes-Casseres, B., Hagedoorn, J., Jaffe, A.B., 2006. Do alliances promote knowledge flows? Journal of Financial Economics. Granovetter, M.S., 1985. Economic action and social structure: the problem of embeddedness. American Journal of Sociology 91, 481–510. Granovetter, M.S., 1995. Coase revisited: business groups in the modern economy. Industrial and Corporate Change 4, 93–140. Griliches, Z., 1990. Patent statistics as economic indicators: a survey. Journal of Economic Literature 28 (4), 1661–1707. Hamilton, G.G., Biggart, N.W., 1988. Market, culture, and authority: a comparative analysis of management and organization in the far east. American Journal of Sociology 94, S52–S94 (Supplement). Hamilton, G.G., Kao, C.S., 1990. The institutional foundation of Chinese business: the family firm in Taiwan. In: Calhoun, C. (Ed.), Comparative Social Research. JAI, Greenwich, pp. 135–151. Hansen, M.T., 1999. The search-transfer problem: the role of weak ties in sharing knowledge across organizational subunits. Administrative Science Quarterly 44, 82–111. Hardle, W., 1990. Applied Nonparametric Regression. Cambridge University Press. Hausman, J., Hall, Z., Griliches, 1984. Economic models for count data with an application to the patents—R&D relationship. Econometrica 52, 909–938. Heckman, J.J., 1979. Sample selection bias as a specification error. Econometrica 47 (1), 153–161. Helfat, C.E., 1994. Know-how complementarities and knowledge transfer within firms: the case of R&D. Management Science 40, 1720–1747. Hill, C.W.L., 1992. Strategies for exploiting technological innovations: when to when not to license. Organization Science 3, 428–441. Hobday, M., 1995. Innovation in East Asia: The Challenge to Japan. Aldershot, Hants, England. E. Elgar, Brookfield, VT, USA. Holland, P.W., 1986. Statistics and Causal Inference. Journal of the American Statistical Association 81 (396), 945–960. Johnson, R.A., Wichern, D.W., 2002. Applied Multivariate Statistical Analysis, fifth ed. Prentice Hall, Upper Saddle River, NJ, 07458. Khanna, T., Gulati, R., Nohria, N., 1998. The dynamics of learning alliances: competition, cooperation, and relative scope. Strategic Management Journal 19, 193–210. Khanna, T., Palepu, K., 1997. Why focused strategies may be wrong for emerging markets. Harvard Business Review 75 (4). Khanna, T., Rivkin, J., 2001. Estimating the performance effects of networks in emerging markets. Strategic Management Journal 22 (1), 45–74. Kock, C.J., Guillén, M.F., 2001. Strategy and structure in developing countries: business groups as an evolutionary response to opportunities for unrelated diversification. Industrial and Corporate Change 10 (1), 77–113. Kodama, F., 1986. Technological diversification of Japanese industry. Science 233 (4761), 291–296.

1503

Kogut, B., 1988. Joint ventures: theoretical and empirical perspectives. Strategic Management Journal 9, 319–332. Lall, S., 2002. Linking FDI and Technology Development for Capacity Building and Strategic Competitiveness. UNCTAD. LaLonde, R., 1986. Evaluating the econometric evaluations of training programs with experimental data. American Economic Review. Lambert, D., 1992. Zero-inflated Poisson regression with an application to defects in manufacturing. Technometrics 34, 1–14. La Porta, R.F., Lopez-de-Silanes, A., Shleifer, 1999. Corporate ownership around the world. Journal of Finance 54, 471–517. Levin, R.C., Klevorick, A.K., Nelson, R.R., Winter, S.G., Gilbert, R., Griliches, Z., 1987. Appropriating the returns from industrial research and development, brookings papers on economic activity. Special Issue On Microeconomics 3, 783–831. Luo, X., Chung, C.N., 2005. Keeping it all in the family: the role of particularistic relationships in business group performance during institutional transition. Administrative Science Quarterly 50, 404–439. Lundvall, B.A., 1998. Why study national systems and national styles of innovations? Technology Analysis and Strategic Management 10, 407–421. Mahmood, I.P., Singh, J., 2003. Technological dynamism in Asia. Research Policy 32, 1031–1054. Mahmood, I.P., Mitchell, W., 2004. Two faces: the effects of business group market share on innovation in emerging economies. Management Science 50 (10). Mahmood, I.P., Mitchell, W. Chung, C.N., 2006. The structure of intra-group ties: innovation in Taiwan Business Groups, Fuqua School Working Paper, Duke University. Miles, R.E., Snow, C.C., 1992. Causes of failure in network organizations. California Management Review Summer, 53–72. Mowery, D.C., Oxley, J.E., Silverman, B.S., 1996. Strategic alliances and interfirm knowledge transfer. Strategic Management Journal 17, 77–91 (special issue). Mullahy, J., 1986. Specification and testing of some modified count data models. Journal of Econometrics 33, 341–365. Nelson, R., 1993. National Innovation Systems. Oxford University Press, New York. Nelson, R.R., Nelson, K., 2002. Technology, institutions, and innovation systems. Research Policy, 31. Nelson, R., Winter, S., 1982. An Evolutionary Theory of Economic Change. Harvard University Press, Cambridge. Numazaki, I., 1986. Networks of Taiwanese big business: a preliminary analysis. Modern China 12 (4), 487–534. Palepu, K., 1985. Diversification strategy. Profit performance and the entropy measure. Strategic Management Journal 6, 239–255. Pisano, G.P., 1990. The R&D boundaries of the firm: an empirical analysis. Administrative Science Quarterly 35, 153–176. Pistor, K., Wellons, A.P., 1998. The Role of Law and Legal Institutions in Asian Economic Development: 1960–1995. Oxford University Press, New York. Rubin, D., 1974. Estimating Causal Effects of Treatments in Randomized and NonRandomized Studies. Journal of Educational Psychology 66, 688–701. Rubin, D., 1977. Assignment to a Treatment Group on the Basis of a Covariate. Journal of Educational Statistics 2 (1), 1–26. Rajan, R., Zingales, L., 1998. Financial dependence and growth. American Economic Review 88, 559–586. Saxonhouse, G.R., 1993. What Does Japanese Trade Structure Tell Us About Japanese Trade Policy? Journal of Economic Perspectives 7 (3), 21–43. Saxenian, A.L., 1999. Silicon Valley’s New Immigrant Entrepreneurs. Public Policy Institute of California, San Francisco. Scherer, F.M., 1982. Inter-industry technology flows in the United States. Research Policy 11, 227–245. Schollhamer, H., 1982. Internal corporate entrepreneurship. In: Kent, C.A., Sexton, D.L., Vesper, K.H. (Eds.), Encyclopedia of Entrepreneurship. Prentice Hall, Englewood Cliffs, NJ, pp. 209–223. Shan, W., 1990. An empirical analysis of organizational strategies by entrepreneurial high-technology firms. Strategic Management Journal 11, 129–139. Singh, K., Mitchell, W., 1996. Precarious collaboration: business survival after partners shut down or form new partnerships. Strategic Management Journal 17, 99–115. Stuart, T.E., Podolny, J.M., 1996. Local search and the evolution of technological capabilities. Strategic Management Journal 17, 21–38, Summer. Teece, D.J., 1986. Profiting from technological Innovation. Research Policy 15, 285–305. Teece, D.J., 1992. Foreign investment and technological development in Silicon Valley. California Management Review 34 (2), 88–107. Vuong, Q.H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica 57, 307–333. Wade, R., 1990. Governing the Market. Princeton U. Press, Princeton. Weinstein, D.E., Yafeh, Y., 1995. Japan’s corporate groups: collusion or competitive? an empirical investigation of Keiretsu behavior. Journal of Industrial Economics 43 (4), 359–376. Williamson, O.E., 1985. The Economic Institutions of Capitalism. Free Press, NY. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. The MIT Press, Cambridge, MA. Yang, C.H., 2008. The effect of strengthening intellectual property rights in NIEs: evidence from Taiwan’s 1994 patent reform. Contemporary Economic Policy 26 (2), 259–275. Zaheer, A., Bell, G.G., 2005. Benefiting from network position: firm capabilities, structural holes, and performance. Strategic Management Journal 26 (9), 809–825.