The Influence of Network Positions on Exploratory Innovation: An Empirical Evidence from China’s Patent Analysis

Abstract

Existing studies find that exploratory innovation requires both access to heterogeneous resources for the establishment of novel knowledge combinations and high-quality cooperation to strengthen the absorption of heterogeneous knowledge. Therefore, structural holes and prominent positions exert critical effects on exploratory innovation. Few existing studies have investigated the influences of the two positions on exploratory innovation simultaneously. This study aims to identify the influence of network positions on exploratory innovation and the contingency mechanism. Basing on joint patent filing relationships identification belong to IPC Green Inventory, R&D collaboration networks in China consisted of 215 firms in the technological field of low-carbon energy were constructed. Negative binomial regression was used to analyse the influences of networks positions on exploratory innovation and the moderating effect of network density. Results reveal that network prominence and structural holes have inverted U-shaped effects on exploratory innovation. The simultaneous occupation of prominent positions and structural holes hinders exploratory innovation. Network density exerts positive and negative effects on the relationships between network prominence and exploratory innovation and between structural holes and exploratory innovation, respectively. The conclusions drawn in this study provide a reference for R&D network structural optimization toward exploratory innovation.

Introduction

The r&d Collaboration network among enterprises in many high-tech industries, such as computers, semiconductors, telecommunications and medicine, has become an important means of realizing innovation (Zhang, 2013). The benefits obtained by organizations from the R&D collaboration network are deemed as a function of network position. For example, in the field of new energy vehicle manufacturing, China’s self-owned brand manufacturing enterprise, namely BYD, has elevated its central position in the R&D network and realized a powerful alliance of key technologies in the field of electric automobiles. This outcome was brought about by the company’s in-depth cooperation with the internationally leading automobile manufacturer Benz and the domestic brand manufacturer Chang’an Automobile. In addition, Chinese domestic brand Geely adopts a loosely connected coalition that spans structural holes through supply chain integration and cooperative R&D. This coalition enabled Geely to obtain diversified, non-redundant information across heterogeneous groups across national borders. The network position is conducive for Geely to accelerate its pace of innovation and gain entrance into the Fortune 500 companies in 2012 (Zhan & Guo, 2015). These examples indicate that two types of network positions, namely prominent positions and structural holes, can exert substantial effects on technological innovation. Thus, its influencing mechanism is worthy of discussion.

Whether and how R&D network positions influence firm innovation has been an ongoing research interest of strategy and organizational studies. Relevant studies mainly concentrate on whether network positions facilitate or constrain innovation outcomes and the contingency mechanism (Dong-Young, 2014; Guan & Liu, 2016; Hahn et al., 2015). According to the current literature, three issues regarding the relationship between network position and firm innovation remain to be further explored. First, the majority of existing studies analyse the direct consequences of network position on innovation performance without further classification of innovation types (Bellamy, Ghosh, & Manpreet, 2014; Guan et al., 2015). Therefore, the question of great importance to contemporary enterprises, i.e., how to effectively use external networks to carry out exploratory innovation, remains to be explained. Recent studies have identified the heterogeneous influences of network position between exploratory and exploitative innovation. The various internal needs of the two types of innovation imply the differentiated influences exerted by network positions (Guan & Liu, 2016). Exploratory innovation highlights novel knowledge acquisition through diversified knowledge searching and requires high-quality cooperative relationships for the absorption of diversified knowledge. The occupation of structural holes can be used to acquire heterogeneous information by associating irrelevant cooperative partners. Meanwhile, network prominence influences resource commitment and information exchange through the quantity and quality of access, resulting in enhanced transboundary transfer of tacit knowledge and consequent knowledge absorption (Koka & Prescott, 2008). Therefore, the relative importance of structural holes and network prominence in exploratory innovation is worthy of in-depth studies. Second, existing studies frequently analyse independent effects of structural holes and network prominence on innovation. However, they neglect the joint effects of the two types of network positions. Network prominence implies the formation of ties with similar others for the alignment of interests. Selecting homogeneous partners facilitates mobilization and pooling of resources for legitimizing and maintaining dominance (Koka & Prescott, 2008). Conversely, the occupation of structural holes implies the establishment of relationships with partners of different statuses (Podolny, 2005). On the one hand, the two types of network positions differ in terms of learning mechanisms and skills. As such, contradiction will exist in the simultaneous pursuit of the two network structures (Koka & Prescott, 2008). On the other hand, the two network positions will generate a complementation due to their mutual reinforcing effects (Zaheer & Soda, 2009). Therefore, investigating the interactive effect between network prominence and structural holes on exploratory innovation is of interest. Third, the inconsistency of existing studies on the impact of network position on innovation may be due to the ignorance of contextual variable. The effects of structural holes and network prominence are dependent on the overall structural characteristics of the network in which the enterprise is embedded. Network density reflects the overall completeness of a network and affects the social capital and knowledge diffusion environment of the node organization. It is an important boundary condition for the investigation of the influence of network positions on exploratory innovation.

According to the foregoing discussion, this study aims to untangle the relationship among prominent positions, structural holes, network density, and exploratory innovation, which can be narrowed down to the following two questions: (1) Do prominent positions and structural holes have different influences on exploratory innovation and is their joint effect negative? (2) How does network density influence this relationship? To this aim, this study constructed R&D collaboration networks based on 2,567 joint patent filing data from nine low-carbon energy technology fields. They were identified under the IPC Green Inventory from 2007 and 2016. In addition, negative binomial regression was used to investigate the influences of network prominence and structural holes on explorative innovation. Furthermore, interaction analysis and three-dimensional figures were combined to explore the moderating effects of network density on this link, so as to reveal what type of network interaction structure is conducive to the realization of exploratory innovation. The hypotheses were tested in the context of low-carbon energy technologies, which comprised nuclear energy, renewable energy, non-renewable energy, and energy conservation technologies (Albino et al., 2014). Different areas of technologies vary in network connection mechanism and cooperative innovation process. This diversity renders rich samples in investigating the mutual influence of network positions and network density. Moreover, low-carbon energy technologies are emerging technologies that are characterized by radical revolution and comparatively higher level of cooperation (Chen, Cheng, Nikic, & Song, 2018; Morcillo-Bellido, Prida-Romero, & Martinez-Belotto, 2018), with which the revolution could cast tremendous effects (Carpio, Martin-Morales, & Zamorano, 2018).

The remainder of this study is organized as follows. The second section presents a theoretical framework and hypothesis development regarding the link between network positions and explorative innovation and the moderating effects of network density. The third section presents the sample data source, network construction method, variable measurement, descriptive statistics, and model specification. The fourth section reports the results of regression and robustness checks. It also discusses favourable and unfavourable network structures to exploratory innovation based on three-dimensional figures of the regression results. Finally, the fifth section summarizes the conclusions and limitations.

Literature Review and Hypothesis Development

Network position is an important indicator of network gains, and its influence on innovation has received continuous attention from scholars (Guan and Liu, 2016; Hahn, et al., 2015; Xue, Zhang, Wang, Skitmore, & Wang, 2018). In particular, the influence of two typical network positions (i.e., network prominence and structural hole) on innovation and the contingency mechanism within it have been a research hotspot in recent years (Dong, McCarthy, & Schoenmakers, 2017; Lee, Wang, & Huang, 2015). Koka and Prescott (2008) argued that network prominence and structural hole differ in terms of learning mechanism and skills, and organizations cannot simultaneously pursue these two network structures. Whether network prominence and structural holes have different influences on innovation remains to be elucidated.

Scholars have explored the above-mentioned problem from two perspectives. On the one hand, scholars point out that the influence of network positions differs between two types of innovation, namely exploratory innovation and exploitative innovation. The former is similar to a disruptive innovation that deviates from the existing knowledge and technology trajectory to explore the niche markets, requiring the understanding and integration of knowledge and technologies to generate a wide range of technology combinations (Danneels, 2002). The latter is similar to increment innovation, which optimizes the existing knowledge and skills for the consolidation of the existing market, requiring a deep understanding and exploitation of the existent technology (Carnabuci & Operti, 2013). Different internal needs of these two types of innovation imply the differentiated influence exerted by network positions. On the other hand, scholars have pointed out that the relationship between network positions and innovation is influenced by contingency factors. Lee et al. (2015) found that the central network position and technological diversity had positive interaction effect on enterprise performance. Dong et al. (2017) found that the influence of network positions on radical innovation is moderated by partner ownership, and a high proportion of private partners can reduce the adverse effects of network prominence on radical innovation. This study discusses the contingency factors from one important dimension of structural embeddedness, namely network density in which findings could reveal the coupling relationship of network structure in the promotion of technological evolution.

Network Prominence and Explorative Innovation

Status refers to a socially constructed, entirely agreed upon and accepted sequence of individuals, organizations, groups or activities in a social system (Washington & Zajac, 2005). Prominent network actors can not only obtain access to abundant network resources through more direct ties but also establish relationships with other organizations of high status because of their superiority in partner selection. These combined effects continuously reinforce their network prominence and release a signal of trustworthiness (Koko & Prescott, 2008). For exploratory innovation, a great number direct collaboration relationships imply more channels to resources and convenience of acquiring complementary knowledge (Gilsing, Nooteboom, Vanhaverbeke, Duysters, & Van den Oord, 2008). In addition, the central network position brings three information advantages, namely accessibility, timeliness and reference, to the node organization (Burt, 1992), thereby enabling node organizations to understand the knowledge distribution, trends and recent developments in the research field and to efficiently evaluate prosperous partners (Wang, Rodan, Frium, & Xu, 2014). Aside from advantages in resources and information, network prominence can bring about cost-saving benefits in terms of information search, supervision, maintenance and transaction, improving the efficiency of knowledge search and information dissemination among organizations (Zhang & Tan, 2014). Therefore, network prominence facilitates firm control in the technology trajectory and direction of new knowledge created in the network by mobilizing resources for its own competitive position (Koko & Prescott, 2008).

However, an increase in number of partners to a certain value brings about information redundancy and diseconomies of scale, which breeds the risks of free riding and involuntary spillover (Wu, 2008). Moreover, the overwhelming information sources increase the difficulty in knowledge absorption. Prominent network actors in R&D networks toward low-carbon energy technologies are usually large state-owned enterprises with monopoly status in the energy supply industry (Ma, Yu, Liu, & Zhang, 2017; Ma, Ye, & Zhang, 2018). These organizations with innate superiority can capture sufficient information in the existing relationships, so they lose the motivation to explore new knowledge. Or they are likely to fall into the existing knowledge paradigm and cognitive framework because the new technological paradigm deviates from the original knowledge base and would impair their authority (Burkhardt & Brass, 1990). Therefore, we propose the following hypothesis:

H1: Network prominence of a firm in the R&D network has an inverted U-shaped effect on its exploratory innovation.

Structural Holes and Exploratory Innovation

Structural hole refers to the gap between non-related parties in the network (Burt, 1992). Organizations occupying structural holes play the role of hub in knowledge transfer and information control as they are in the middle of two organizations without direct contact (Burt, 2004). These organizations are usually advantageous in terms of information, control and independent R&D. With respect to information superiority, these organizations can acquire heterogeneous, non-redundant and timely knowledge and information from two ends of structural holes (Burt, 2004), thereby facilitating them in recombining the heterogeneous knowledge and information into the generation of new knowledge. The control advantage is mainly from the bridge gains of the organization, which plays the role of knowledge dissemination channel and information intermediary (Zhang & Tan, 2014). The advantage of independent R&D means that organizations occupying many structural holes are not easily constrained by the codes of conduct and thinking paradigms within closely contacted groups. Consequently bringing position is conducive to the cultivation of innovation consciousness of the organization (Wang et al., 2014).

However, excessively non-redundant ties would generate certain costs to the bridging organizations. The direct cost comes from the acquisition of non-redundant information, which includes the processing and integration of diverse information. The indirect cost is reflected in the shift of time and effort from absorbing diverse knowledge to continuously adapt knowledge pools from non-relevant fields (Gilsing et al., 2008). Therefore, the over-occupation of structural holes hinders the node organization’s absorption of diverse knowledge and consequently leads to a negative impact on its exploratory innovation. Accordingly, we propose the following hypothesis:

H2: Structural holes of a firm in the R&D network have an inverted U-shaped effect on its exploratory innovation.

Interaction Between Network Prominence and Structural Holes

Scholars hold two contrary opinions over the interactive influence between structural holes and network prominence on innovation. Some believe that network prominence and structural holes are not an either-or relationship but a complementary one. Zaheer and Soda (2009) argued that prominently positioned organizations in a network tend to occupy more structural holes. First, prominently positioned organizations typically have superior knowledge reserve resources. Hence, establishing an individual network with many structural holes is necessary to reduce knowledge leakage risks because structural holes inhibit information transmission in the network. Second, prominently positioned organizations are capable of signing exclusive cooperative agreements with cooperative partners, which facilitate the formation of structural holes. Third, prominently positioned organizations usually have prestige and power, which attract isolated entities in the network and thus increase the formation of structural holes. Stam and Elfring (2008) found that if prominently positioned enterprises occupy certain structural holes at the same time, they will have an enhanced ability to overcome information asymmetry and integrate seemingly irrelevant elements into innovative elements.

Other scholars doubt the abovementioned arguments and posited that if prominently positioned enterprises simultaneously pursue structural holes, their innovation capabilities will be impeded. Perry-Smith (2006) contended that diversified information brought about by structural holes distract attention from existing research work. As such, the benefits of prominently positioned enterprises from structural holes are limited. Zhang (2013) observed that when a prominently positioned enterprise simultaneously occupies many structural holes, its cooperative partner will worry about being exploited. This doubt will abate the partner’s willingness for relationship commitment and knowledge sharing. Based on the previous discussion, structural holes and network prominence may exert a positive or negative interactive influence on exploratory innovation. Thus, the following hypothesis is proposed:

H3: Network status and structural holes can exert an interactive influence on exploratory innovation.

Moderating Effect of Network Density

Network density is the density of node connections in a network structure, characterized by the number of ties in the network and the completeness of the network as a whole. From the perspective of social capital, the increase in the network density is conducive for the nodes in establishing the trust mechanism, which can enhance the cohesion of cooperative groups and the degree of knowledge sharing (Gulati, 1998). From the perspective of knowledge diffusion, high density can improve the efficiency of knowledge transfer and promote the absorption of common norms and knowledge (Zeng & Wen, 2015). Therefore, network density affects the social capital and knowledge diffusion environment of node organizations, which then influence the relationship between network positions and exploratory innovation.

When network density is low, the number and complexity of collaborative ties in the network are low, and the information and resource acquired by prominent node organizations are limited. Simultaneously, given the uneven distribution of the network power in the sparse network (Zhang & Tan, 2014), few organization with more homogeneous partners and high status are more likely to lock in the existing innovation path, which weakens the positive effect of the network prominence on exploratory innovation. Low network density has a negative effect on the speed and range of knowledge diffusion, thereby highlighting the advantage in diverse knowledge acquisition brought by structural holes and the bridge gains from the role of information intermediary. Thus, the positive impact of structural holes on explorative innovation could be strengthened.

When network density is high, a large number of connections between nodes improve the network connectivity and transmission efficiency, accelerate the social capital flow and the willingness of knowledge sharing among organizations, thereby improving the efficient absorption and integration of diverse knowledge by prominent network actors (Schilling & Phelps, 2007; Zhao, Wang, & Zheng, 2016). High network density not only reduces the information screening and absorption limitation brought by network prominence but also eases the innovation vision limit (Koko & Prescott, 2008). High density means that nodes in the network have frequent connections and most organizations are connected directly or indirectly with other organizations, thereby reducing the heterogeneity of information and resources and weakening the information value and control advantage obtained by structural holes (Burt, 2000). Accordingly, we propose the following hypotheses:

H4: Network density positively moderates the relationship between network prominence and exploratory innovation.

H5: Network density negatively moderates the relationship between structure holes and exploratory innovation.

Methodology

Sample Selection

Sample was collected from patent data pertaining to low-carbon energy technology in China from 2007 to 2016. The identification of low-carbon energy technology was based on the IPC green inventory disclosed by WIPO,¹ with which the IPC codes were sorted following the work of Albino et al. (2014). Nineteen subclasses of low-carbon energy technology that belonged to renewable energy, non-renewable energy, energy conservation and storage technology were generated. Patents were retrieved from the Baiten database, which is a widely used and fully functional patent information platform in China.² Patent data mining is due to its representation of firm’s inventions (Roca-Gonzalez, Vera-Lopez, & Rodriguez-Bermudez, 2018). Specifically, joint filing patent is not only a comprehensive and objective reflection of the results of cooperation among organizations but also a commonly used indicator for constructing R&D collaboration networks in existing studies (Guan & Liu, 2016; Suk, Lee, & Jeong, 2016). For sample filtering, entries that have jointly filed patents among individuals or between individuals and organizations were removed, leaving only patent entries among organizations (Zhang & Chen, 2013). Jointly filed patents between parent and subsidiary firms were eliminated to rule out influence of non-innovation motives, such as performance needs and cost sharing.

A sample of 491 organizations with 2567 jointly filed patents in the field of low-carbon energy technologies was acquired. The remaining 15 subclasses of low-carbon technology were further merged into 9 categories, namely wind energy, low-carbon construction, waste-to-energy incineration, green lighting, energy saving and storage, bioenergy, nuclear energy, solar energy and fuel cells. The merging of sample is for the purpose of yielding sufficient jointly filing patents in single R&D network construction without losing sample. We extracted only firm samples and finally kept 251 companies in our sample with 1789 jointly filed patents from 2007 to 2016. The industry distribution of our sample is shown in Figure 1.

The jointly filed patents were converted one by one into two-mode matrices. Figure 2 exhibits the rationale of R&D network construction, where Figure (a) denotes the generation of a two-mode matrix and Figure (b) stands for a one-mode network. The conversion of (a) to (b) indicates that whenever at least one collaborative patent exists among filing organizations over a given period, a link exists with filing organizations being nodes and collaboration times being link values, that is, one-mode matrix. The collaborative relationship has the property of time continuity. Thus, existing literature generally uses 3–5 years as time windows in generating R&D networks (Stuart, 2000; Zhao & Zheng, 2013). Considering the emerging property of low-carbon energy technologies and the comparatively small sample of joint patent filing organizations, we selected 5 years as the time windows (years t-2, t-1, t, t+1 and t+2) to generate the adjacency matrix of year t (Ma et al., 2017). We merged the years 2007 and 2008 into one due to the small sample quantities in these years. Thus, jointly filed patent data from 2007 to 2016 was used to generate R&D collaboration networks from 2010 to 2014.

Figure 1.

Source: Baiteng database.

Figure 2.

Source: Author’s illustration of network construction.

Variable Measurement and Descriptive Statistics

Table 1 summarizes the variable measurement of the model.

Table 1

Measurement of Variables and Data Sources

Variable Name	Variable Description
Dependent variable
Explorative innovation (Y)	Number of low-carbon energy technology patents a firm successfully filed in year t within patent classes, in which the former four digits of the IPC codes have not been active in the five years prior to the granting year.
Independent variables
Network density (D)	Number of existing links in a network divided by all possible links among networking organizations calculated for 45 adjacency matrices of R&D networks that comprised 5 time horizons multiplied by 9 low-carbon technology subclasses.
Network prominence (NP)	In accordance with the measurement of Ahuja, Polidoro, & Mitchell (2009), Bonacich’s eigenvector centrality was calculated for the 45 adjacency matrices of R&D networks that comprised 5 time horizons multiplied by 9 low-carbon technology subclasses. The Bonacich’s eigenvector centrality measures the quantity and status of collaboration partners, thus providing a consistent measure of network prominence. This measure was standardized to compare values across R&D networks among 9 low-carbon technology subclasses.
Structural holes (SH)	In accordance with the measurement of Yang, Lin, and Peng (2011), Burt’s (1992) constraint measure of structural holes was calculated for 45 adjacency matrices of R&D networks that comprised 5 time horizons multiplied by 9 low-carbon technology subclasses. The expression is $C_{i j} = {(p_{i j} + \sum_{q} p_{i q} p_{q j})}^{2}; q \neq i, j$ , where p_ij denotes the investment proportion of firm i in the relationship of firm j in the network, and $\sum_{q} p_{i q} p_{q j}$ is the total amount of indirect relationship of firm i in the network. For reverse encoding, the constraint measures of structural holes were multiplied by minus one. This measure was standardized to compare values across technology R&D networks among 9 low-carbon technology subclasses.
Control variables
R&D intensity (RD)	Stock of patents granted to a focal firm in the past four years prior to a given year t.
Explorative learning capability (A1)	Number of explorative patents granted to a focal firm in the previous year of a given year t.
Exploitative learning capability (A2)	Difference between a focal firm’s total patents granted in year t-1 and the number of explorative patents granted in year t-1.
Ownership (O)	Dummy variable that equals to 1 if a firm is state-owned and 0 otherwise
Sector (Sector)	Sector dummy
Time (Year)	Year dummy

Source: Ownership data were collected from enterprise websites, and sector belongings were identified from corresponding enterprise information with sector classification by the National Bureau of Statistics.³ Data for the remaining variables were collected from the Baiten database.

Table 2 lists descriptive statistics of the variables. Since our measure of RD intensity is the stock of patents granted to a focal firm in the past four years prior to a given year t, this number is non-negative integer and varies from firm to firm. In our sample, the minimum patent stock of a focal firm is 0 and the maximum patent stock of a focal firm is 5357. This descriptive statistics indicates heterogeneity in the firms’ RD intensity, and consequently offer validity of being control variables. For the regression analysis, we have dealt with this extreme value problem with winsorizing.

Table 2

Descriptive Statistics of Variables

Variables	N	Mean	Std	Min	Median	Max
D	757	0.080	0.0380	0.033	0.08	0.180
NP	757	1.494	1.671	0.147	0.98	8.754
NP2	757	5.018	13.03	0.022	0.961	76.64
SH	757	1.532	4.025	0	0	23.98
SH2	757	18.524	77.033	0	0	574.88
NP*SH	757	5.136	20.289	0	0	209.896
D*NP	757	0.145	0.235	0.005	0.069	1.576
D*SH	757	0.141	0.393	0	0	3.143
A1	757	7.844	14.25	0	2	101
A2	757	55.91	229.6	0	2	2221
RD	757	190.0	646.7	0	23	5357
O	757	0.563	0.496	0	1	1
Y	757	11.76	98.52	0	8	305

Source: Author’s calculation.

Notes: D is density; NP and NP2 are the measures of network prominence and its square; SH and SH2 are the measures of structural hole and its square; NP*SH is the product of NP and SH; D*NP is the product of D and NP; D*SH is the product of D and SH; A1 and A2 are the metrics of Explorative learning capability and Exploitative learning capability respectively; RD is RD intensity calculated from patent stock; O is ownership dummy; Y is explorative innovation calculated by explorative patents granted for a focal firm.

Furthermore, the correlation matrix presented in Table 3 indicates that the correlation coefficients for the quadratic terms and the interactions terms are very high. Therefore, we standardized the quadratic and interaction terms before entering into the models. Second, we conducted a VIF test of the independent variables. The average value of VIF is 2.56, and the maximum value is 7.36, which is less than 10, so multicollinearity problem between the independent variables is not a serious concern according to Peter (2008) and Chen (2014).

Table 3

Correlation Matrix of Variables

	D	NP	NP2	SH	SH2	NP*SH	D*NP	D*SH	A1	A2	RD	O
D	1
NP	0.396**	1
NP2	0.370**	0.942**	1
SH	0.121**	0.424**	0.373**	1
SH2	0.078**	0.372**	0.344**	0.934**	1
NPSH*	0.167**	0.577**	0.592**	0.829**	0.842**	1
DNP*	0.545**	0.943**	0.851**	0.337**	0.286**	0.506**	1
DSH*	0.226**	0.502**	0.478**	0.746**	0.885**	0.894**	0.462**	1
A1	0.062*	0.088**	0.060*	0.306**	0.262**	0.222**	0.063*	0.270**	1
A2	0.019	0.090**	0.052	0.278**	0.239**	0.186**	0.049	0.240**	0.536**	1
RD	0.015	0.088**	0.051	0.275**	0.240**	0.188**	0.045	0.234**	0.494**	0.624**	1
O	0.120**	0.124**	0.142**	0.158**	0.132**	0.154**	0.135**	0.147**					0.156**	0.101**	0.112**	1

Source: Authors’ calculation.

Note: * p<0.1, ** p<0.05, *** p<0.01 (Two-tailed test).

Model Specification

The dependent variable explorative patent is non-negative count variables with which OLS regression leads to biased estimates of coefficients (Chen, 2014). Poisson or negative binomial models are appropriate for count dependent variables. The selection between these two models is based on the variance of dependent variables. Table 2 indicates that the variance of explorative patents is greater than the mean value, which is indicative of ‘over-dispersion’. Thus, the negative binomial regression model is selected. We first conducted fixed effect panel data model to see the result of F tests and reported it in the fourth last line in Table 4. Since the p-value of F tests all equals to zero, which indicates that fixed effects panel model is superior to pooled regression. Then, we decided between fixed effects and random effects panel data models. Given that our sample data is panel data ranging from 2007 to 2016, the Hausman test was used to determine whether random effects or fixed effects panel regression is appropriate model. The results of the Hausman tests in Table 4 suggest that all models fail to decline the null hypotheses at significant levels. In addition, fixed effects model produces biased estimates when sample duration is short. Therefore, the random effects negative binomial regression model was selected as basic model (Guan & Liu, 2016). Furthermore, selection between negative binomial models and zero-inflated negative binomial models is necessary because 20% of explorative patents are zero. We performed the Vuong test and found that the Vuong statistics of all regressions are below 1.96 in Table 5, indicating neither of the models is preferred (Vuong, 1989). Therefore, we first reported the estimation results of negative binomial models and then applied zero-inflated binomial models for robustness check because panel data option in STATA 12.0 is unavailable for zero-inflated negative binomial models (Guan & Liu, 2016).

Table 4

Random Effects Panel Negative Binomial Regression Results for Exploratory Innovation

Variables	Dependent Variable: Explorative Innovation Y
Variables	M1	M2	M3	M4	M5
NP		0.078***		0.021***	0.041***
		(2.79)		(2.65)	(2.80)
NP2		−0.011***		−0.023**	−0.031***
		(−3.61)		(−2.35)	(−2.64)
SH			0.011***	0.016***	0.022***
			(3.21)	(3.71)	(3.53)
SH2			−0.056***	−0.016**	−0.031***
			(−3.97)	(−2.31)	(−2.64)
NP*SH				−0.020**
				(−1.74)
D					6.816*
					(1.79)
D*NP					2.593***
					(3.00)
D*SH					−0.799***
					(−3.50)
A1	0.013***	0.012***	0.012***	0.012***	0.012***
	(4.77)	(4.76)	(4.57)	(5.02)	(4.44)
A2	0.056***	0.031***	0.026***	0.019***	0.019***
	(5.32)	(4.01)	(5.01)	(4.92)	(4.32)
RD	0.132**	0.153**	0.176**	0.173**	0.143**
	(2.36)	(2.34)	(2.16)	(3.23)	(2.09)
O	0.466***	0.484***	0.443***	0.456***	0.457***
	(3.82)	(4.00)	(3.56)	(3.81)	(3.52)
Year dummy	√	√	√	√	√
Industry dummy	√	√	√	√	√
F statistics	50.49	51.93	51.84	52.52	56.90
Wald chi2	113.69	118.88	156.34	156.70	191.66
log-likelihod	−2298.16	−2296.26	−2297.462	−2283.830	−2041.086
LR test	45.73***	54.28***	55.56***	57.49***	65.59***

Source: Authors’ calculation.

Note: * p<0.1, ** p<0.05, *** p<0.01 (T statistics in the parentheses).

Results and Discussion

To authenticate the proposed hypotheses, we used STATA 12.0 to perform random effects negative binomial panel regressions for explorative innovation. The regression results were reported in Table 4. Models M1 to M5 represent the basic model with only control variables, models of single key explanatory variables, and models with key explanatory and interaction variables. Overall, the signs of coefficients are stable across models. The LR statistics in every model declines the null hypothesis of Poisson distribution at 1% significant level, thereby indicating the appropriateness of our model specification of negative binomial regression. For negative binomial regressions, the goodness of fit statistic is Wald chi square. The Wald chi square in model M1 to M5 passes significance tests, which indicates a good level of model fitness. Sector and year effects were controlled by dummy variables.

Result Analysis

The independent influence of network prominence and structural holes on exploratory innovation was tested and the regression results were reported in models M2 and M3, respectively. H1 predicts that network prominence exerts an inverted U-shaped effect on explorative innovation. The coefficients of NP and its quadratic terms in Model 2 and Model 5 support H1 at p < 0.01, thus confirming H1. This result indicates that the initial increases in network prominence of the firm in R&D collaboration network tend to favour its explorative innovation. However, when network prominence reaches a certain value, further increases are likely to hinder explorative innovation. This result is in line with that of Zeng and Wen (2015). At the parabola threshold, the value of network prominence is 3.55, which is much higher than the sample mean value of network prominence at 1.494. This result implies that the majority of firms in the R&D collaboration networks are in the phase of rewarding from the enhancement of network prominence. Linking with either many organizations with heterogeneous knowledge or very important partners in the R&D networks are beneficial to the explorative innovation of firms. H2 contends that structural holes exert an inverted U-shaped effect on explorative innovation. The coefficients of SH and its quadratic terms in Model 3 and Model 5 exhibit the expected signs and are significant at p < 0.01, thus supporting H2. Contrary to network prominence, the estimated value of structural hole for exploratory innovation at the threshold is 0.098, which is lower than its mean sample value of 1.532. This result indicates that a large fraction of firms in the R&D networks are in the phase of excessive occupation of structural holes, which leads to a negative effect on explorative innovation because firms are less likely to acquire non-redundant knowledge. Model M4 reports the results on the interactive effects of network prominence and structural holes. NP*SH is significantly negatively correlated with exploratory innovation (β = -0.020, p < 0.01). Therefore H3 is supported. The LR statistics of Model 2, Model 3 and Model 4 are higher than that in of Model 1, which implies that the inclusion of key independent variables increases goodness of fit. In terms of the magnitude of influence, a comparison of coefficients between NP and SH indicates that the former position exerts a stronger positive effect on exploratory innovation than the latter one does.

The result pertaining to the moderation effect of network density on the link between network positions and exploratory innovation was reported by model M5. LR statistics in Model 5 is higher than those of other models, which indicates that the inclusion of interaction terms further improves goodness of fit of the model. The coefficient of D*NP in Model 5 supports H4 at p < 0.01. Central firms in dense R&D networks are superior in terms of technology and scale, and accumulate considerable resources, heterogeneous knowledge and capacity to develop cutting-edge technologies. This drives the exploration of novel technologies and therefore exerts a positive impact on explorative innovation. H5 proposes that network density negatively moderates the link between structural holes and explorative innovation. The coefficients of D*SH in Model 5 supports H5 at p < 0.01. This result is consistent with the results of Gilsing et al. (2008) and Zhang and Tan (2014). A high network density weakens the information advantages of structural holes because most organizations in the R&D networks have direct or indirect relationships with one other, which hinders bridging firms from acquiring of heterogeneous information.

Explorative learning capability and exploitative learning capability are positively related to explorative innovation, thus confirming that explorative innovation and exploitative innovation are interdependent events. The coefficients of state ownership are significantly positive in all models. This result implies that compared with non-state-owned enterprises, state-owned enterprises usually have scale and financial superiorities and are also subject to more supervision from government and society. Therefore, state-owned enterprises would have stronger capability and motivation to implement low-carbon energy technological innovation.

Furthermore, three-dimensional graphs and two mode matrixes were combined to indicate the favourable and undesirable circumstances for explorative innovation. Figure 3 displays the interaction between network density and network prominence on explorative innovation and the inferred corresponding relationship between interaction types and explorative innovation. Figure 3 shows that a higher level of exploratory innovation appears in central firms in dense R&D networks, whereas lower level ones appear in central firms in sparse R&D networks. Therefore, whether a central firm plays a positive or negative role in the R&D collaboration networks depends on the context of network cohesion. Figure 4 depicts the interaction between network density and structural holes on explorative innovation and the inferred corresponding relationship between interaction types and explorative innovation. Figure 4 demonstrates that the lower level of exploratory innovation occurs in firms occupying more structural holes in dense networks, whereas a higher level of exploratory innovation appears in firms with more structural holes in sparse networks. Accordingly, when network cohesion is low, occupation of structural holes are conducive to the improvement of exploratory innovation, whereas high degree of connectivity and aggregation of R&D network diminishes the information and resource advantages of bridging firms. Figure 3 and Figure 4 jointly confirm the moderating effect of network density on the link between network positions and explorative innovation.

Figure 3

Source: Authors’ calculation.

Figure 4

Source: Authors’ calculation.

Robustness Test

We ran several additional robust tests to verify the reliability of the results. For potential omitted variable bias, we introduced a classification dummy of low-carbon energy technologies into the model to control the effect of a single technology network, thus avoiding the results influenced by a certain type of R&D network. For model specification, we modelled explorative innovation by zero-inflated binomial models following Guan and Liu (2016). For variable measurement, we took account of weights in measuring explorative innovation owing to the highly skewed distribution of patent data. The generally adopted metrics in measuring patent value are forward citations, patent family size, year of renewal and the number of claims in patent application (Guan & Liu, 2016). We used the number of claims in patent application as the weights of our dependent variables due to data availability. Tong and Frame (1994) argued that the more the claims in a patent application, the higher of its technical quality and the acknowledged innovation value. In Table 5, Wald chi squares in model M6–M10 passes significance tests which are indicative of acceptable model fitness. The results show that the signs and significance of the estimated coefficients of relevant variables are similar to those in Table 4, thereby proving the consistency of our previous findings.

Table 5

Zero-inflated Binomial Regression for Patent Claim-Weighted Explorative Innovation

Variable	Dependent Variable: Patent Claims Weighted Explorative Innovation Y
Variable	M6	M7	M8	M9	M10
NP		0.073***		0.021***	0.060***
		(2.81)		(2.63)	(2.96)
NP2		−0.009***		−0.002**	−0.001***
		(−4.46)		(−2.31)	(−2.87)
SH			0.012***	0.011***	0.011***
			(.70)	(3.85)	(3.26)
SH2			−0.057***	−0.028**	−0.029***
			(−4.17)	(−2.37)	(−2.89)
NP*SH				−0.034*
				(−1.80)
D					6.012*
					(1.78)
D*NP					2.023***
					(2.97)
D*SH					−0.510***
					(−9.14)
A1	0.018***	0.018***	0.018***	0.010***	0.018***
	(33.90)	(33.70)	(33.98)	(30.42)	(32.31)
A2	0.006***	0.007***	0.006***	0.005***	0.005***
	(3.32)	(3.35)	(3.34)	(4.37)	(3.36)
RD	0.135**	0.134**	0.134**	0.133**	0.130**
	(2.15)	(2.18)	(2.16)	(2.23)	(2.01)
O	0.376***	0.376***	0.371***	0.360***	0.369***
	(11.38)	(11.39)	(10.98)	(10.81)	(10.89)
Tech-class dummy	√	√	√	√	√
Year dummy	√	√	√	√	√
Industry dummy	√	√	√	√	√
log-likelihod	−2882.492	−2877.669	−2877.211	−2876.883	−2870.354
Wald chi2	3912.49	3937.75	3966.68	3984.90	4096.14
Vuong test	1.74	1.	1.35	1.18	1.56

Source: Author’s calculation.

Note: * p<0.1, ** p<0.05, *** p<0.01 (T statistics in the parentheses).

Discussion

According to the empirical results, H1, H2, H3, H4 and H5 were all verified. Therefore, this study established a framework to explain the influence of network position on exploratory innovation. On the basis of the empirical results, the following advices were proposed for enterprises to exploit position gains in improving exploratory innovation output. The policy implications is presented in Figure 5.

Figure 5

Source: Author’s discussion.

First, the study results verify H1 and H2. That is, structural holes and network prominence have an inverted U-shaped effect on exploratory innovation. This result is in line with that of Guan and Liu (2016). However, it is inconsistent with the conclusion that structural holes do not affect exploratory innovation. This result indicates the impact of structural holes is substantial if its occupation is controlled at a moderate level. Moreover, structural holes and network prominence differ in their influencing degrees and phases. A comparison of regression coefficients between network prominence and structural holes shows that network prominence exerts a greater influence on exploratory innovation. This result indicates that advantages in terms of resources and information acquisition, cooperative partner selection, and cost saving brought about by network prominence in the current R&D network toward low-carbon energy technology are more important to exploratory innovation than the advantage of heterogeneous knowledge brought about by structural holes. In terms of threshold comparison, the peak values for network prominence and structural holes that were estimated by exploratory innovation are higher than their mean values (threshold values are 3.55 and 0.098, respectively, and mean values of the samples are 1.494 and 1.532, respectively). Enterprises should appropriately elevate their network status by increasing direct and indirect network relations or cooperating with partners with high influence in the collaborative network. Given that the efficiency of enterprises in leveraging structural holes in collaborative R&D remains at a low level, one possible reason is that enterprises fail to attach importance to the improvement of their absorption capability when cooperating with a heterogeneous partner. Consequently, their internal capability cannot match the demand for the advantage utilization of structural holes. This finding is in line with the propositions of Cohen and Levinthal (1990). Therefore, enterprises should improve their knowledge acquisition, digestion, conversion and application capacities to match the internal requirements for structural holes.

Second, the results of this study support H3. That is, enterprises’ simultaneous occupation of prominent positions and structural holes is not conducive to exploratory innovation. This result is consistent with that of Kola and Precott (2008). This result indicates that diversified information brought about by structural holes will distract attention from the existing research work and inhibit the benefits acquired by prominently positioned enterprises from structural holes. In addition, when prominently positioned enterprises occupy structural holes at the same time, their collaboration partners will worry about being exploited, which lessens the partners’ relationship commitment and knowledge sharing. Therefore, enterprise managers should dynamically realize a balance between two network positions and overcome the loss of another position advantage through partner selection. For instance, enterprises pursuing high network status can select partners with diversified technologies to carry out in-depth cooperation. This option can guarantee enterprises’ acquisition of non-redundant resources and avoid the increase of knowledge processing load and cooperative friction brought about by the simultaneous occupation of structural holes.

Third, the results in this study further verify H4 and H5. That is, network density positively moderates the relationship between network prominence and exploratory innovation. However, it negatively moderates the relationship between structural holes and exploratory innovation. These results imply that the coupling of network density and position creates favourable and unfavourable network structures for exploratory innovation. In R&D collaboration networks with low densities, prominently positioned firms have a negative effect on exploratory innovation. These central enterprises are mostly state owned and faced with a high degree of system obstacle and path dependence in green transformation. Therefore, the government should use regulatory policies to force such enterprises to actively develop low-carbon technologies. In R&D collaboration networks with low densities, bridging firms have a positive effect on the level of exploratory innovation. The government can promote the construction of strategic alliance and industry association to strengthen knowledge hub function and expand the knowledge diffusion scope of bridging firms. In R&D networks with high densities, bridging firms exert an inhibitory effect. The government can adopt guidance policies to improve bridging firms’ enthusiasm in the acquisition of heterogeneous information. The government can guide bridging firms to undertake the information hub task in research, testing, demonstration and marketization of low-carbon technology, rather than restricting them in the R&D stage. It can also guide bridging firms to participate in trans-regional cooperation projects and international cooperation projects. In this manner, the channels for acquiring heterogeneous information are widened. In R&D networks with high densities, central firms have a positive effect. The government can adopt incentive and ensuring policies to strengthen the leading and demonstration role of central firms. Meanwhile, the infrastructure–supply policy is an important policy that spans through the emergence to development stages of low-carbon energy technology R&D networks. It provides an indispensable basis for the development of R&D networks due to its fundamental role in the supply of innovation elements.

Conclusion

Based on the panel data of 215 enterprises, which had joint patent filings in the fields of low-carbon energy technologies from year 2007 to 2016, this study empirically analysed the relationship between network position and exploratory innovation and explored the moderation effects of network density. The following conclusions were drawn.

First, an inverted U-shaped relationship exists between a firm’s network prominence and level of exploratory innovation. The majority of firms are in the stage where the enhancement of network status is conducive to exploratory innovation. This result indicates that most firms in our sample R&D networks can continue to increase its collaborative ties with other influential organizations to benefit from information and resource advantages. In addition, an inverted U-shaped relationship exists between structural holes and exploratory innovation. The majority of firms are in the stage where the over-occupation of structural holes hinders the generation of exploratory innovation output. These findings suggest that bridging position results in several advantages, such as information control, resource manipulation, opportunity perception and risk response for exploratory innovation. However, the occupation of structural holes, to a certain degree, leads to redundant information, which negatively affects exploratory innovation.

Second, the simultaneous pursuit of network prominence and structural holes hinders enterprises’ exploratory innovation. For prominently positioned enterprises, diversified information brought about by structural holes disperses the node organizations’ attention from existing research work. Therefore, the benefits obtained by prominently positioned enterprises from structural holes are limited. For enterprises that occupy many structural holes, their cooperation with prominently positioned enterprises can generate the potential of being exploited and thus weaken their willingness for relationship commitment and knowledge exchange.

Third, the density of R&D networks positively moderates the influence of network prominence on exploratory innovation and negatively moderates the influence of structural holes on exploratory innovation. Therefore, single-level analysis is insufficient to reflect the overall influence of network structure on innovation. The comprehensive consideration of actor position and global network structure can shed light on network structural optimization towards technological breakthroughs.

Our study has several limitations that can be addressed in future studies. First, only network density, which proxies the external knowledge diffusion environment was considered as our contingency variable. Future studies can exploit contingency variables from the perspectives of enterprises’ internal capabilities or resources in the investigation of the nexus between network positions and exploratory innovation. Second, only joint filing patent data were used to construct R&D networks due to data availability. However, joint filing patent data are limited in the reflection of the overall R&D collaboration activities of organizations. Future studies can collect a comprehensive dataset, which includes additional data on patent purchase, transfer and cross-licensing for empirical analysis.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Acknowledgements

The authors gratefully acknowledge the support by the National Social Science Fund of China under the funding program code 17BJL031.

Notes

References

Ahuja

, Polidoro

, & Mitchell

(2009). Structural homophily or social asymmetry? The formation of alliances by poorly embedded firms. Strategic Management Journal, 30(9), 941–958.

Albino

, Ardito

, Dangelico

R. M.

, & Petruzzelli

A. M.

(2014). Understanding the development trends of low-carbon energy technologies: A patent analysis. Applied Energy, 135, 836–854.

Bellamy

M. A.

, Ghosh

, & Manpreet

(2014). The influence of supply network structure on firm performance. Journal of Operations Management, 32(6), 357–373.

Burkhardt

M. E.

, & Brass

D. J.

(1990). Changing patterns or patterns of change: The effects of a change in technology on social network structure and power. Administrative Science Quarterly, 35(1), 104–127.

Burt

R. S.

(1992). Structural holes: The social structure of competition. Cambridge, UK: Oxford University Press.

Burt

R. S.

(2000). The network structure of social capital. Research in Organizational Behavior, 22, 345–423.

Burt

R. S.

(2004). Structural holes and good ideas. American Journal of Sociology, 110(2), 349–399.

Carnabuci

, & Operti

(2013). Where do firms’ recombinant capabilities come from? Intraorganizational networks, knowledge, and firms’ ability to innovate through technological recombination. Strategic Management Journal, 34(13), 1591–1613.

Carpio

, Martin-Morales

, & Zamorano

(2018). Environmental and economic effects of using renewable energy in residential thermal installations according to 2030 targets: Case study in the province of Granada (Spain). DYNA, 93(5), 474–477.

10.

Chen

(2014). Advanced econometric economics and STATA applications. Beijing, China: Higher Education Press.

11.

Chen

J.-D.

, Cheng

S.-L.

, Nikic

, & Song

M.-L.

(2018). Quo vadis? Major players in global coal consumption and emission reduction. Transformations in Business & Economics, 17(1), 112–132.

12.

Cohen

W. M.

, & Levinthal

D. A.

(1990). Absorptive capability: A new perspective on learning and innovation. Administrative Science Quarterly, 35(1), 128–152.

13.

Danneels

(2002). The dynamics of product innovation and firm competence. Strategic Management Journal, 23(12), 1095–1121.

14.

Dong-Young

(2014). Understanding supplier structural embeddedness: A social network perspective. Journal of Operations Management, 32(5), 219–231.

15.

Dong

J. Q.

, McCarthy

K. J.

, & Schoenmakers

W. W. M. E.

(2017). How central is too central? Organizing interorganizational collaboration networks for breakthrough innovation. Journal of Product Innovation Management, 34(4), 526–542.

16.

Gilsing

, Nooteboom

, Vanhaverbeke

, Duysters

, & Van den Oord

(2008). Network embeddedness and the exploration of novel technologies: technological distance, betweenness centrality and density. Research Policy, 37(10), 1717–1731.

17.

Guan

J.-C.

, Zhang

J.-J.

, & Yan

(2015). The impact of multilevel networks on innovation. Research Policy, 44(3), 545–559.

18.

Guan

J-C.

, & Liu

(2016). Exploitative and exploratory innovations in knowledge network and collaboration network: A patent analysis in the technological field of nano-energy. Research Policy, 45(1), 97–112.

19.

Gulati

(1998). Alliances and networks. Strategic Management Journal, 19(4), 293–317.

20.

Hahn

M. H.

, Lee

K. C.

, & Lee

D. S.

(2015). Network structure, organizational learning culture, and employee creativity in system integration companies: The mediating effects of exploitation and exploration. Computers in Human Behavior, 42, 167–175.

21.

Koka

B. R.

, & Prescott

J. E.

(2008). Designing alliance networks: the influence of network position, environmental change, and strategy on firm performance. Strategic Management Journal. 29(6), 639–661.

22.

Lee

C.-Y.

, Wang

M.-C.

, & Huang

Y.-C.

(2015). The double-edged sword of technological diversity in R&D alliances: Network position and learning speed as moderators. European Management Journal, 33(6), 450–461.

23.

Liu

F.-Z.

, Wu

D.-L.

, & Ma

R.-K.

(2015). Patent technology licensing and innovation performance of licenses firms: The moderating effects of three different types of proximity. Science Research Management, 36(4), 91–100.

24.

, Yu

Y.-S.

, Liu

S.-S.

, & Zhang

Y.-R.

(2017). Dynamics of waste-to-energy incineration R&D collaboration networks: A social network analysis based on patent data. Geosystem Engineering, 20(2), 59–70.

25.

, Ye

J.-M.

, & Zhang

Y.-R.

(2018). Can firm exploit economic gains from eco-innovation? An empirical investigation of listed companies in China. Transformations in Business & Economics, 17(2), 33–53.

26.

Morcillo-Bellido

, Prida-Romero

, & Martinez-Belotto

(2018) Sustainability: How energy service company (ESCO) contribute. DYNA, 93(4), 385–390.

27.

Perry-Smith

J. E.

(2006). Social yet creative: The role of social relationships in facilitating individual creativity. The Academy of Management Journal, 49(1), 85–101.

28.

Peter

(2008). A guide to econometrics. Hoboken, NJ: Wiley-Blackwell Press.

29.

Podolny

J.M.

(2005). Status signals: A sociological study of market competition. Princeton, NJ: Princeton University Press.

30.

Roca-Gonzalez

J. L.

, Vera-Lopez

J. A.

, & Rodriguez-Bermudez

(2018). Analysis of patent #US2014/0319274A1: A case study of simulations for new designs review. International Journal of Simulation Modelling, 17(3): 405–418.

31.

Schilling

M. A.

, & Phelps

C. C.

(2007). Inter-firm collaboration networks: The impact of large-scale network structure on firm innovation. Management Science, 53(7), 1113–1126.

32.

Stam

, & Elfring

(2008). Entrepreneurial orientation and new venture performance: The moderating role of intra- and extraindustry social capital. The Academy of Management Journal, 51(1), 97–111.

33.

Stuart

T. E.

(2000). Interorganizational alliances and the performance of firms: A study of growth and innovation rates in a high-technology industry. Strategic Management Journal, 21(8), 791–811.

34.

Suk

, Lee

S.-Y.

, & Jeong

Y. S.

(2016). A survey on the impediments to low carbon technology investment of the petrochemical industry in Korea. Journal of Cleaner Production, 133, 576–588.

35.

Tong

X.-S.

, & Frame

J. D.

(1994). Measuring national technological performance with patent claims data. Research Policy, 23(2), 133–141.

36.

Vuong

Q. H.

(1989). Likelihood ratio tests for model selection and non-nested hypotheses. Econometrica, 57(2), 307–333.

37.

Wang

C.-L.

, Rodan

, Frium

, & Xu

X.-Y.

(2014). Knowledge networks, collaboration networks, and exploratory innovation. Academy of Management Journal, 57(2), 484–514.

38.

Washington

, & Zajac

E. J.

(2005). Status evolution and competition: Theory and evidence. Academy of Management Journal, 48(2), 282–296.

39.

W-p.

(2008). Dimensions of social capital and firm competitiveness improvement: The mediating role of information sharing. Journal of Management Studies, 45(1), 122–146.

40.

Xue

, Zhang

, Wang

, Skitmore

, & Wang

(2018). Analyzing collaborative relationships among industrialized construction technology innovation organizations: A combined SNA and SEM approach. Journal of Cleaner Production, 173, 265–277.

41.

Yang

, Lin

, & Peng

M. W.

(2011). Behind acquisitions of alliance partners: Exploratory learning and network embeddedness. Academy of Management Journal, 54(5), 1069–1080.

42.

Zaheer

, & Soda

(2009). Network evolution: The origins of structural holes. Administrative Science Quarterly, 54(1), 1–31.

43.

Zhan

, & Guo

(2015). Less-endowed firms’ strategy in multiple asymmetric alliances: A comparative double-case study. Collected Essays on Finance and Economics, 197(8), 89–95.

44.

Zhang

G.-X.

(2013). How to management network prominence and structural holes in R&D alliances? A MOA model perspective. Management World, 11, 89–100, 129.

45.

Zhang

G.-P.

, & Chen

X.-D.

(2013). Analysis on patent quality in new energy technology field — A case study on wind and solar energy technologies. R&D Management, 25(1), 73–81.

46.

Zhang

H.-J.

, & Tan

J.-S.

(2014). Alliance network and firm innovation performance: A hierarchical perspective. Management World, 3, 163–169.

47.

Zhao

, & Zheng

X.-J.

(2013). Network-based clustering, connectivity and firm knowledge innovation: An empirical research on alliance relation network in 10 high-tech industries in China. Science of Science and Management of S. & T., 34(3), 23–32.

48.

Zhao

, Wang

, & Zheng

X.-J.

(2016). Impact of network vicinity and geographical proximity to knowledge transfer performance. Science Research Management, 37(1), 128–136.

49.

Zeng

D.-M.

, & Wen

J.-Y.

(2015). The influences of R&D collaboration network centrality, knowledge distance to firm’s dual-innovation. Chinese Journal of Management, 12(10), 1479–1486.