Disentangling Learning Network Dilemma: A Small-World Effect in a Globalized World

Abstract

During this pandemic, Research and Development (R&D) firms were faced with the challenge to engage in collaborative networks to immediately find the cure for coronavirus. However, the closed and local model of the innovation ecosystem causes the innovation process carried out by a single laboratory to be slow and ineffective. We study how R&D firms should configure the open innovation ecosystem network for optimal collaborative learning. We argue that value creation in collaborative learning can be influenced by configuring structural connections and relational cohesion in a network of inter-organizational R&D collaborations. A model based on a combination of two network configurations, namely, inter-network connections and intra-network cohesion, was tested on 204 R&D collaborations from the pharmaceutical industry. Our study found an interaction effect between inter-network connections and intra-network cohesion on knowledge acquisition performance. Furthermore, when seeking optimal knowledge transfer, increasing investment commitment in R&D collaboration is more effective than extending the duration of a relationship. This study contributed a dynamic model of collaborative learning by testing the complementary effects between structural and relational configurations in the external and internal firm’s innovation ecosystem for sustainable knowledge acquisition performance.

1. Introduction

The sources of new technology, knowledge, skills, and capabilities have spread internationally and are assimilated in different locations and ecosystems [1,2]. However, in the midst of a current global pandemic situation, mobility between countries is restricted to prevent the spread of the virus. R&D firms need to be more strategic in managing their international R&D collaboration. The study of multi-agent cooperation in an interorganizational network (ION) [3,4] plays an essential role in the discussion of international R&D alliances, and it emphasizes the importance of building an effective network configuration [5,6,7] as part of a company’s global R&D strategy.

Research on knowledge network configuration has been divided into two rival streams. One school of thought supports closed network structures, emphasizing the importance of control, coordination, and increased cooperation in tight and closed networks [8,9]. Another group focuses on the importance of open networks and the role of bridging ties to increase the scope of accessible information and knowledge resources [10,11].

In the context of fighting the global pandemic, both closed and open network typologies have limitations. Maintaining bridging ties in open networks can increase opportunities to explore combined knowledge, but it hinders the benefits of control and coordination through an efficient closed network configuration.

Although many scholars suggested the benefit of open innovation to recombine knowledge from external sources [12,13,14], to what extent this benefit will be inhibited by network typologies remains unclear.

The answer to this question will partly depend on the firm’s capability to incorporate its R&D network configuration as a truly knowledge-intensive enterprise (KIE) [12]. When establishing an inter-organizational R&D network, firms are confronted with choices of whether to expand the connection across different innovation ecosystems or enhance the cohesion in the existing network ecosystem.

I use a small-world research lens, which assumes that bridging external ties outside the innovation ecosystem cluster (inter-network connection) and internal relational strength inside the innovation ecosystem cluster (intra-network cohesion) should be treated in two different constructs that complement each other. According to the small-world theory [15], both external bridging ties and relational cohesion must coexist to positively influence innovation [16]. In our research setting, small-world networks are identified by a network structure that has a high level of relational cohesion inside the firm’s innovation ecosystem and a high number of bridging connections outside the firm’s external innovation ecosystem.

I focus this study on two research questions. First, to what extent does bridging ties across different networks (inter-network connection) affect the knowledge acquisition from the innovation ecosystem? Second, how do the two types of relational configurations in the internal network (intra-network cohesion) moderate the relationship between external bridging ties and the firm’s capacity to acquire knowledge from the network of innovation ecosystem?

To answer those questions, I conduct patent-based research on 204 R&D collaborations in the pharmaceutical industry. This study contributes in several ways. First, this study provides an important opportunity to advance the understanding of knowledge network configuration in inter-organizational R&D. Research on evaluating the role of bridging ties and structural embeddedness in inter-organizational R&D networks remains inconclusive [17], as the previous study only focuses on the attributes associated with the structural configuration of the network and neglecting the relational configuration both inside and outside innovation ecosystem.

Second, I advance previous studies examining the complementary role of the structural and relational configuration of networks [18,19,20] by further detailing two different arrangements of intra-network cohesion: Alliance duration and investment commitment. I integrated the leading theories of the Resource-Based View [21], Transaction Cost Economics [22], and Social Network Theory [10,23] to explain how some of these relational cohesion arrangements work together as systems with structural configuration to improve knowledge acquisition performance.

Third, it is hoped that this research will contribute to a deeper understanding of examining the effect of structural and relational configurations on the knowledge network by differentiating these configurations based on the external and internal firm’s innovation ecosystem. Using this approach, I can shed more light on the complexities of managing the alliance portfolio mix [17] both at the ecosystem level and individual alliance level of analysis.

The remainder of the paper is organized as follows: The next section provides an overview of the literature relevant to the main research question. In Section 3, I introduce our research methodology, variables, and data collection process. The analysis and empirical findings are presented in Section 4. In Section 5, I discuss these findings and make some comparisons with the existing literature. Finally, this paper draws conclusions and several avenues for future research direction.

2. Literature Review

2.1. The Paradox of Network Structure

In the last few decades, scholars agree that Knowledge-Intensive Enterprises (KIE) need to integrate multiple sources of knowledge from external sources and foster innovation [24,25].

The study of network theory provides useful insights for understanding the role of networks in promoting collaboration and innovation performance [4,6,7,26].

The position of an actor in a network gives individuals or organizations the ability to access information and obtain other resources [10,27] and at the same time, they can share risks with external partners [28].

However, in the context of learning networks, networking configuration presents some paradoxes for companies. Scholars have debated which structural configuration is beneficial for innovative collaboration.

The first paradox is based on the understanding that although firms need to access new knowledge from their partners, companies naturally want to keep their own knowledge from imitation. To what extent should the firm’s R&D be embedded in the innovation network ecosystem? Engaging in a deeply embedded network provides an opportunity to access new knowledge but also means firms are risking their competitive advantage.

The second paradox stems from the fact that studies examining the best network configurations in innovation networks remain inconclusive. One school of thought supports open network structures to increase the scope of accessible knowledge resources [10,11]. Other groups emphasize the importance of trust, coordination, and increased cooperation in tight and closed networks [8,9].

Most debates among social network theorists are polarized on the aspect of structural configurations consisting of features such as network density, centrality, between, and the degree of redundancy in innovation networks [5,29,30], and neglect the interaction between the effect of relational dimensions inside and outside the innovation ecosystem.

In this study, I separated the structural and relational configuration into two distinct dimensions that exist in the innovation ecosystem network. I differentiate structural and relational configurations in inter-organizational R&D based on external and internal innovation ecosystem networks.

Previous studies suggested that optimal firms’ learning is influenced by two configurations of network embeddedness simultaneously: Structural and relational embeddedness [18,31]. Both structural and relational embeddedness are important to a firm’s innovative performance, although these two configurations have different roles. The structural configuration relates to the position of the company in the overall innovation ecosystem network, which determines the opportunities to access new knowledge. On the other hand, the relational configuration is related to the quality of the relationship, which influences the flow of knowledge among network members [31].

In this study, I extend the context of open innovation in inter-organizational R&D networks [2,32,33,34] by further distinguishing the structural configuration in terms of bridging the inter-network connection (outside innovation ecosystem), and relational configurations are considered as building intra-network cohesion (inside innovation ecosystem). The following section will elaborate further.

2.2. Inter-Network Connection and Knowledge Acquisition Performance

From the resource-based perspective [21], the motivation for companies to engage in strategic alliances is to access new valuable resources from their partners [35,36]. Scholars who integrate a resource-based viewpoint perspective with social network theory believe that a company’s specific network position is one of the basic resources for a company’s performance [5,11,37]. Networking gives companies an approach to accessing information and obtaining other resources [10,27] from their innovation ecosystem [38,39].

The term innovation ecosystem was first used by Moore [40] to describe an interconnected network of firms that coevolve capabilities around a shared set of technologies or knowledge and work cooperatively to develop new products.

Keeping this definition, the study of the structural configuration of R&D firms is concerned with the position of the firm in the innovation ecosystem network and how this position influences the potential to reach the various data circulating in that network [6,41]. R&D firms can benefit from their innovation ecosystem network if they are successful in increasing the range of information that can be accessed through bridging structural holes in their ecosystem networks [11]. Knowledge-intensive enterprises (KIE) can achieve higher innovative performance by maintaining intermediaries [42] or bridging ties with non-redundant actors [10].

In this study, I focus on observing the benefits of bridging external ties outside the innovation ecosystem network, suggesting that the existence of second-hand brokerage or connections between actors to whom one is only connected indirectly increases networking benefits more dramatically [43].

I highlight external sources of knowledge and observe how bridging inter-network connections determines knowledge acquisition performance. Bridging the inter-network connection refers to the position of R&D at the intersection of a heterogeneous network ecosystem with diverse competencies, which brings the benefit of increasing the scope of accessible information and access to leveraging distant knowledge resources [11].

Previous research has found that the benefits of bridging ties are not always persistent in every circumstance. Several explanations regarding the contradictory consequences of bridging ties are discussed in various papers. Experts argue that the effect of bridging ties on innovation performance depends on many factors such as relational attributes [20], the focal company’s attributes [36], and characteristics of the alter [7,44].

So far, the role of bridging ties on learning performance has been investigated in scattered studies. I am interested in examining the extent to which bridging inter-network connections affects an R&D firm’s capacity to acquire knowledge from its innovation ecosystem network. The mainstream social network studies suggest that bridging ties generally have a positive influence on innovation [10,11]. Therefore, I formulate the following hypothesis:

Hypothesis 1.

Bridging inter-network connections will affect knowledge acquisition performance positively.

2.3. Intra-Network Cohesion and Knowledge Acquisition Performance

Instead of using a resource-based viewpoint, the relational configuration in inter-organizational R&D has been investigated using transaction cost theory [22,45]. Under this theory, the motivation for the R&D firms to form an innovation ecosystem network is driven by the need to reduce the costs and complexities associated with contracting out uncertain markets. A learning network in this sense is considered a form of intermediate hybrid between market and hierarchy [46].

Intra-network cohesion can affect knowledge acquisition performance in general, since problems of complexity and uncertainty to integrate critical resources in a competitive market will be partially resolved by increasing commitment and repeated interactions among companies in the network ecosystem. If the structural configuration captures the position properties and the number of useful links in the network, the relational configuration addresses the quality of the relationships among members in the innovation ecosystem network.

Intra-network cohesion, in this study, is reflected in the intensity of the relationship based on investment commitments among partners in an ecosystem network (e.g., joint ventures, outsourcing, licensing, or franchise alliances) or a longer relationship duration [29].

In the context of knowledge transfer in the innovation ecosystem network, strong relational cohesion is considered positive for efficient transfer processes [47,48] because the movement of new ideas from other units is influenced by a strong degree of trust and willingness to be involved in the knowledge transfer process [49].

However, this premise seems to contradict the strength of the weak ties paradox [23]. This theory argues that weak relational ties are built by loosely connected relationships with more actors who are preferable to provide a wider range of information. Because each actor can operate on a different network, weak ties have the advantage of offering access to new and different information.

Scholars who integrate social network theory and transaction cost economics are unable to reach a consensus on issues of relational cohesion and performance [44,50]. In the context of the inter-organizational R&D ecosystem, strong relational cohesion is considered a double-edged sword [17,51]. With strong relational cohesion, R&D firms can develop an intense understanding that positively enhances the company’s innovative capabilities. Unfortunately, this also reduces the chances of discovering new information as much of the information circulating in tightly ties systems is redundant.

Hansen [19] has shown that the relationship between relational cohesion and learning performance may differ from case to case depending on the firm’s preferred conditions.

Built from this argument, I propose two types of relational cohesion, namely, alliance duration and investment commitment, as two different conditions that correlate to knowledge acquisition performance.

I find that the elaboration and identification of each quality have been largely ignored by social network theorists. For that reason, I propose two hypotheses to observe how these two different types of relational cohesion affect knowledge acquisition performance:

Hypothesis 2a.

Intra-network cohesion in terms of the longer duration of an alliance will affect knowledge acquisition performance positively.

Hypothesis 2b.

Intra-network cohesion in terms of stronger investment commitment will affect knowledge acquisition performance positively.

2.4. Joint Consideration: Small-World Perspective

To understand the knowledge network configuration in inter-organizational R&D, it is not enough to focus on observing the structural and relational configuration on a separate basis. Important elements of the interaction of structural and relational configuration both outside and inside the innovation ecosystem networks must also be taken into account. Structural configuration in terms of the quantity of the bridging innovation ecosystem network is useful for the acquisition of unique information [10], whereas a strong relational configuration is useful in addressing the quality of knowledge being transferred [19,52].

I take an integrative view of the small-world theory [15], which provides promising clues to resolve debates of structural and relational configurations on the social network. Small-world networks in learning are characterized by two simultaneous conditions: Strong relational ties in the internal innovation ecosystem and the existence of external bridging ties with other innovation ecosystem networks [16]. In our study, small-world networks refer to a kind of innovation ecosystem with high cohesion, as measured by the relational dimension, while the average number of intermediates to connect any actors in different network ecosystems is relatively short, as measured by the average path length [53].

Figure 1 illustrates our constructs of inter-network connections and intra-network cohesion. A focal firm, F, will develop an inter-network connection when it acts as a brokerage and has ties with two other companies from different network clusters, and the two do not connect. At the same time, the focal company may have links with other companies in the same network cluster to create intra-network cohesion.

By integrating inter-network connections and intra-network cohesion, I not only adopted a small-world perspective in the context of inter-organizational R&D but I also elaborated on the joint mechanism of structural and relational configurations for optimizing knowledge acquisition performance. Our explanation builds on the previous discussion on exploration and exploitation for knowledge-seeking activity abroad. Kuemmerle [54] also drew on a dichotomous set of motivations for FDI in R&D, home-base-exploiting (HBE) and home-base-augmenting (HBA) motivations, while Ambos and Schlegelmilch [55] used the terminology “capability exploiting” and “capability augmenting” to address the contingent mandate of an R&D unit.

A knowledge-based view of firms also suggests distinct mechanisms of knowledge stock accumulation and knowledge transfer as two capabilities that represent a source of competitive advantage in the R&D investment [12,53,56].

I argue that the inter-network connection (structural) and intra-network cohesion (relational) affect the sequential processes of the accumulation of knowledge stocks and knowledge transfer. Knowledge stock is important because it provides a company with a foundation for core competencies [56], while knowledge transfer is essential for enabling the dynamic capabilities of the company [57] by facilitating the company to expand, refine, and modify its knowledge stock.

The accumulation of the knowledge stock requires conditions that facilitate the efficient identification process of knowledge from various sources outside the network ecosystem. Leveraging bridging ties with non-redundant ties will allow R&D firms to identify new information worthy of a variety of resources.

However, although a rich open network with bridging ties increases opportunities to explore new knowledge, it hinders the benefits of exploiting existing knowledge through coordination and cooperation in a closed ecosystem network.

In the following hypotheses, I propose the knowledge acquisition performance of the inter-network connection outside the innovation ecosystem will be improved by considering the moderating factor of intra-network cohesion inside the innovation ecosystem.

Hypothesis 3a.

The longer the duration of a relationship between focal firms and their network members, the stronger the relationship between a firm’s bridging inter-network connection and its subsequent knowledge acquisition performance.

Hypothesis 3b.

The stronger the investment commitment between focal firms and their network members, the stronger the relationship between a firm’s bridging inter-network connection and its subsequent knowledge acquisition performance.

3. Research Methods

3.1. Data Collection Process

The objective of this study is to understand the knowledge network configuration in inter-organizational R&D. For this purpose, I conducted network studies based on the 204 R&D alliances in the pharmaceutical and biotech industry, which are technology-intensive industries where the exchange of knowledge and dependence on knowledge acquisition is high.

R&D firms in these industries are mostly engaged in innovative ecosystem networks to create synergies through the cross-fertilization of learning and to manage the risks and costs associated with the internationalization of R&D processes.

In situations where companies are embedded in a different network ecosystem, R&D firms often interact with unfamiliar knowledge that has been created in completely different environments.

Research on examining structural network configuration requires data availability at the population level. For this reason, the data used in this study are entirely public data, namely patent registration data at the U.S. Patent office available online. For the relational dimension of the alliance, we used available alliance data from the Thomson Reuters Recap database between 2000 and 2004.

The R&D network configuration was extracted from the recap database, which provides data on a comprehensive alliance of pharmaceutical companies around the world. I set the time window between 1 January 2000, and 31 December 2004, and extracted all alliance information between those dates.

In total, I included 204 focal firms and 967 partners. We chose the timeframe for the study using data from 2000 to 2004 to replicate the data collection process as suggested by Rogbeer et al., [32], with additional considerations. First, the level of data recency is a factor that is not very relevant in this study because we believe that the relationship between the configuration of the network structure and the learning outcome is not influenced by when or where the context of the research was carried out.

Second, from 2002 to 2004, the world also experienced the Severe Acute Respiratory Syndrome (SARS) outbreak, which has a similar virus strain to the current Coronavirus disease. With these arguments, we assumed that using data from 2000 to 2004 is sufficient to represent the current condition.

Third, instead of taking the static point of view on these particular years, we were more interested in using longitudinal sampling methods to understand the dynamic changes over some period of time, therefore we assumed this approach would provide better insight to understand the cause-and-effect relationship, regardless of the time selection window.

In order to measure the actual knowledge acquisition performance across the alliance network, I took all the relevant patent dataset information from the USPTO (US Patent and Trademark Office), including the citation profiles of each patent. The USPTO requires applicants to include relevant previous patent citations in their patent information to ensure the objectivity and consistency of the information. Therefore, patents and their citations provide a practical setting for measuring the cumulative learning of network members, and they also imply the integration and recombination of resources derived from multiple sources, locations, and organizational positions [58]. By incorporating both structural and relational dimensions, which span across multiple levels of analysis, I elucidated the complex mechanism of learning optimization through different configurations in the network ecosystem.

3.2. Variables

3.2.1. Dependent Variable

The main dependent variable was studying the results of knowledge acquisition performance from the firm’s innovation ecosystem networks. I defined knowledge acquisition performance as the cross-fertilization of knowledge interchange between companies as a consequence of interactions when there is a commitment to collaborate in a network of alliances. I counted the total number of cross-citations of the focal company’s patents with their partners, divided by the number of patents generated each year. This variable summarizes the company’s actual knowledge creation and knowledge acquisition from multiple sources. I labeled this variable CROSSLEARNING; this increase in number is an indication of the extent to which focal companies on learning and acquiring new technologies from their learning network [59].

As an analogy, this measure is comparable to citation behavior in scientific journals. When two scholars work together, they begin to share their knowledge. The increasing number of overlapping patent references indicates a build-up of knowledge stock and knowledge acquisition from network partners. This variable is calculated as three cumulative reference years that overlap after the alliance is formed.

The previous study uses qualitative case studies to understand value creation in knowledge-intensive enterprises [3]. However, this method is not valid for analyzing long-term trends in understanding knowledge value creation. Longitudinal sampling methods in our study have several advantages over cross-sectional studies in providing details of changes over time. I adapt the data processing method as suggested by Rogbeer et al. [32] to better understand the knowledge acquisition performance in the inter-organizational R&D ecosystem.

3.2.2. Independent Variable

Our independent variables were the external bridging ties (as a proxy for the inter-network connection) and the internal relational ties (as a proxy for intra-relational cohesion). The interaction between these two variables was observed in our model.

For the first independent variable, I measured the extent to which actors find themselves embedded in the external network using Burt’s broker criteria. Using UCINET software, I calculated the number of pairs external to the network (BRIDGE) brokerage pairs not directly connected to the focal firm. The number of external bridging ties was not the same for all organizations, and the extent to which focal firms act as bridging ties can vary. By counting the number of partners not directly connected, I evaluated the potential of the focal firm to access new information from the entire alliance ecosystem network.

For the second independent variable, Rowley et al. [20] stated that cohesion is associated with the definition of tie strength as proposed by Granovetter [23]. Tie strength refers to the combination of the amount of time, emotional intensity, intimacy (mutual confiding), and mutual service that characterize a tie [23]. In this study, I followed Gilsing and Duysters’ [29] interpretation of relational embeddedness, which classifies two different measures of tie strength: Duration (DURATION) and investment commitment (COMMITMENT). I measured the average number of times in terms of accumulative months for DURATION and the proportion of equity–non-equity alliances versus the total number of alliances as our indicator for COMMITMENT [60].

3.2.3. Control Variables

To control for unobserved heterogeneity, three variables were introduced as controls in the model. Based on previous studies, I considered several aspects such as R&D expense (RDEXPENSE), the number of employees (EMPLOYEE), and past patents (PASTPATENTS). I collect general information from the BioScan and Recap databases.

4. Results

Descriptive Statistics

Descriptive statistics and variable correlations are presented in

Table 1. Descriptive statistics.

		Mean	Std. Dev.	Min	Max	Correlation
						1	2	3	4	5	6	7	8	9
1	CROSS LEARNING	12.2	46.85305	0	754.3	1
2	BRIDGE	0.46	0.0966208	0	0.5	0.0078	1
3	DURATION	8.27	10.09756	0	89.78	−0.0590	0.0913	1
4	COMMITMENT	0.42	0.1752647	0	1	0.0425	−0.0248	−0.0433	1
5	BRIDGE × DURATION	3.91	4.881081	0	43.64	−0.0565	0.1783	0.9798	−0.0599	1
6	BRIDGE × COMMITMENT	0.19	0.0829157	0	0.5	0.0312	0.5086	0.0254	0.5998	0.0735	1
7	RDEXPENSES	221	647.4362	0	5176	0.0397	0.0941	−0.1960	0.0187	−0.1893	0.0827	1
8	EMPLOYEES	342	4780.698	0	72,400	−0.0180	0.0227	−0.0533	−0.0405	−0.0519	−0.0273	0.2726	1
9	PASTPATENTS	9.69	15.68352	0	163	−0.0023	0.1101	−0.1462	0.0194	−0.1373	0.0985	0.2554	−0.0325	1

Notes: Values above 0.16 are significant at p < 0.05.

. The mean value, standard deviation, and correlation matrix for all variables are presented in this table. In three years, the sample companies had accumulated an average of 12.2 cross-citations per patent. I found that the internetwork bridging connection (BRIDGE) has a low correlation with DURATION (0.0913) and COMMITMENT (−0.0248).

I used a non-linear estimator of the negative binomial regression model, which is typically used in patent studies with a limited number of outputs (in our case, overlapping references) and a high number of zeros [61]. I conducted our analysis using the following empirical model:

CROSSLEARNING = f (BRIDGE, COMMITMENT, DURATION, CONTROLS)

The results of the regression analysis are summarized in

Table 2. Analyzes of cross-learning network: Negative binomial regression models. (DV: Network partner cross citation counts).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7
	Base	H1	H2A	H2B	H3A	H3B	All
BRIDGE		0.4063598	0.6227885	0.2648514	0.9298271 *	1.496469 *	2.00849 *
			(1.46)	(0.62)	(1.91)	(1.76)	(2.24)
DURATION			0.0157299 ***		0.0186378		0.017009
			(3.2)		(0.9)		(2.6)
COMMITMENT				0.4060065 **		0.3819233	0.4020775
				(1.81)		(0.78)	(−1.69)
BRIDGE × DURATION					0.0740816 ***		0.0699795
					(1.68)		(1.61)
BRIDGE × COMMITMENT						2.113444 *	2.117666 *
						(1.84)	(1.82)
RDEXPENSES	0.0001913 ***	0.0001882 ***	0.0001557 ***	0.0001948 ***	0.0001527 ***	0.0001977 ***	0.0001626
	(3.52)	(1.3)	(1.59)	(3.59)	(2.75)	(3.66)	(2.95)
EMPLOYEE	−0.0000302 ***	−0.0000303 ***	−0.000031 **	−0.000031 **	−0.0000311 **	−0.0000315 **	−0.0000323 ***
	(−1.99)	(−2)	(−2.04)	(−2.05)	(−2.05)	(−2.08)	(−2.13)
PASTPATENT	0.0057363 ***	0.0055596 ***	0.004529 **	0.0057595 ***	0.0044285 **	0.0058656 ***	−0.0047641 **
	(2.84)	(2.74)	(2.19)	(2.86)	(2.14)	(2.92)	(2.33)
INTERCEPT	−1.643274 ***	−1.828162 ***	−1.78343 ***	−1.97919 ***	−1.918362 ***	−2.092763 ***	−2.188599 ***
	(−32.69)	(−9.22)	(−8.94)	(−17.58)	(−8.48)	(−5.32)	(−5.28)
χ2 for covariates	34.21	34.92	43.66	38.06	45.41	41.63	51.57
(d.f)	5	5	5	5	5	5	5
N	204	204	204	204	204	204	204

Notes: z values shown in parentheses below the regression coefficients * p < 0.1; ** p < 0.05, *** p < 0.01; one-tailed test.

.

In Model 1, only control variables are included to provide the basic model for this study. This model elucidates that all control variables have a significant effect on knowledge acquisition.

Model 2 includes variables bridging external ties (BRIDGE). Hypotheses H2A and H2B are tested in model 3 (for COMMITMENT) and model 4 (for DURATION). The effect of the interaction between structural and relational embeddedness is observed in Model 5 and Model 6, while the full model of this study is represented in Model 7.

When the external bridging variable was treated partially, this variable failed to achieve a significant result in Model 2 (Hypothesis 1). Similar results have been found in previous studies, which suggest that a positive outcome of bridging ties should depend on several other factors that should be included in the analysis [11,62]. There is no clear relationship between external bridging ties and learning on a stand-alone basis. The insignificant findings shown in Model 2 suggest that observing solely bridging ties is still not sufficient to explain optimal learning performance. The results from the observation of two types of intra-network cohesion (H2A and H2B) broadly support our hypothesis. Both DURATION (Model 3) and COMMITMENT (Model 4) are significant and positive for increasing knowledge acquisition performance.

However, one of the main findings in this study is the influence of the interaction between inter-network connection and intra-network cohesion variables. Our third hypothesis examines the performance impact of the interaction between these two variables. Both BRIDGE × DURATION and BRIDGE × COMMITMENT are statistically significant (Models 5 and 6). The interaction effect between these two variables also increases the knowledge acquisition coefficient drastically compared to individual variables. The positive coefficient estimates in Models 5 and 6 show that firms increase their capacity to learn as they jointly consider the relational and structural configurations in the network.

What stands out in the table is that there was a significant difference between the increasing investment commitment and the duration of the alliance. A comparison of the two models reveals that increasing investment commitment has contributed to the increase in knowledge acquisition performance (with a much higher regression coefficient than duration). These results indicate the need to examine external bridging and internal relational configurations, particularly investment commitment, when analyzing the best network configuration for optimal knowledge acquisition performance.

5. Discussion

The main focus of this study was to examine the attributes associated with structural configuration and relational configuration both inside and outside innovation ecosystem networks.

This study adds to the literature on how R&D firms can optimize their network configuration for optimal knowledge acquisition performance, especially in the context of inter-organizational R&D networks. I observe how the different types of intra-network relational cohesion, i.e., alliance duration and investment commitment, influence knowledge acquisition performance in different ways.

The increasing number of non-redundant ties (bridging ties) can increase the opportunity to access new knowledge. However, having too many non-redundant ties is disadvantageous for the focal firm as it increases the complexity of integrating that diverse knowledge. Previous research has cited this occurrence by saying that as bridging ties extend to a certain point, the costs and negative outcomes may outweigh the benefits [5,63,64,65]. I found that maintaining strong relational intra-network cohesion could be one of the important strategic decisions for focal companies in mitigating this problem.

Our study shows that internal relational cohesion in innovation ecosystem networks affects knowledge acquisition performance differently. Different internal relational settings within the innovation ecosystem network will be either more or less effective to optimize the knowledge acquisition by bridging the ecosystem network inter-organizational R&D alliances. As shown in Table 2, the increasing coefficient for COMMITMENT is greater than DURATION. From these results, I can conclude that increasing investment commitment is more effective than extending the duration of a relationship.

Furthermore, I would argue that the premise of isolating the role of relational strength as a cause of optimal knowledge acquisition is, by itself, not a sufficient explanation. Therefore, I use a more holistic approach to jointly consider relational and structural embeddedness to achieve completeness in observing the learning phenomenon of the network.

Adopting the small-world theory to include both external bridging ties and internal relations in the innovation network can help resolve debates about structural and relational configurations. Distinguishing levels of network embeddedness inside and outside the network ecosystem will help us understand the mechanisms through which the two embeddedness configurations influence knowledge acquisition. Being embedded in strong internal relationships in the network and bridging external networks outside the innovation ecosystem networks can provide a significant advantage, allowing focal companies to acquire knowledge efficiently from the cluster while, at the same time, accessing new ideas from distant sources [66,67].

A comparison of models 5, 6, and 7 shows that the interaction effect between relational and structural configurations does provide a statistical advantage in explaining knowledge acquisition performance. Structural and relational configurations show a strong significant influence compared to significance on a stand-alone basis. These results confirm the small-world theory, which says that both internal relational strength and external bridging ties must complement each other in improving learning performance. A higher degree of small worldness means that the benefits of acquiring new knowledge from external resources and assimilating knowledge from an internal ecosystem network are also high. The small-world characteristics of the alliance network configuration have a positive effect on the cross-fertilization of learning. From the above findings, we conclude that the two different network specifications of connection and cohesion indicate the necessity of both features in inter-organizational R&D to optimize the knowledge acquisition process.

Going beyond these findings, I predict the effects of co-interaction between structural and relational embeddedness as well as an alternative solution to how firms optimize absorption capacity [68] and internalize new knowledge from outside the organization. Lane, Koka, and Phatak [69] suggest a more comprehensive definition of absorptive capacity construction based on their in-depth study of the literature: “Absorption capacity is the ability of firms to utilize their knowledge externally through three sequential processes: (1) exploratory learning, recognizing and understanding potentially valuable knowledge outside the organization, (2) transformative learning, or assimilating new knowledge, (3) exploitative learning, to re-create new knowledge and to enhance commercial outcomes”.

Figure 2 shows that strengthening the inter-network connection and intra-network cohesion affects the sequential process of exploratory learning and exploitative learning. Maintaining external bridging connections is beneficial for increasing the knowledge stock as the likelihood of the focal companies accessing a variety of knowledge also increases. However, above a certain level, the knowledge dissemination and information flow of alliance partners will decrease as the complexity of integrating multiple sources of knowledge increases, so partners who are over-diversified lead to less innovative results for the focal firm [5]. To reduce this problem, companies need to maintain strong relational cohesion in their alliance network. Our findings suggest that increasing relational ties in internal networks has a positive effect on learning, through mechanisms of trust accumulation and cohesion in the network of alliances. Moreover, when pursuing optimal knowledge transfer, increasing investment commitment in R&D collaboration is more beneficial than extending the duration of the relationship.

Focal firms can receive more benefits from bridging access to a wide range of information from the external network ecosystem and more effectively assimilating knowledge among internal network members. In this way, companies can explore new opportunities in the inter-network environment while, at the same time, exploiting beneficial knowledge sources available via the intra-network. Because these two features are important, these two conditions must be balanced to maintain organizational learning [70].

These results also extend to a stream of the empirical literature on organizational ambidexterity about how the company should manage the tension between exploration and exploitation (March 1991). The proposition to reconcile the balance between exploration and exploitation is intuitively interesting and has been widely discussed in academic fields [71,72]. This paper provides empirical evidence to investigate the mutual consideration of inter-network connections and intra-network cohesion as one possible way of achieving ambidexterity learning in the context of alliance inter-organizational R&D. These findings have significant implications for understanding how structural and relational dimensions in network configuration complement each other and should help to improve predictions of the impact of increased investment commitment on knowledge acquisition performance.

6. Summary and Conclusions

As coronavirus strains mutate rapidly in different corners of the world, building learning networks is believed to be one of the most effective ways to develop and integrate knowledge from various sources. However, how R&D firms should configure the learning network structure for optimal collaborative learning remains unclear.

Over-relying on inter-network connections through cross-border R&D alliances in this global pandemic situation provides some dilemmas. Bridging networks across countries may increase access to various possible external sources. However, the excessive bridging connections may not be beneficial for firms since the benefit of establishing a partnership with partners from various countries and technological backgrounds cannot be materialized without having substantial relational commitment among firms. This lack of relational linkage could potentially hinder the knowledge transfer process. This study reveals that the learning benefit from inter-network connections can be affected by configuring the intra-network cohesion inside the region.

Our study suggests that bridging ties appear to be a simple construction as they have a special role in increasing access to various possible sources. However, excessive bridging ties may not be beneficial for the company, because the benefits of forming partnerships with partners who come from various cognitive backgrounds cannot be realized without substantial relational commitment between firms. A lack of relational relationships has the potential to hinder the knowledge transfer process. I suggest that the learning benefits of external bridging inter-network ties can be influenced by configuring intra-network cohesion.

This study has several contributions to management practice. First, it is important to understand how companies manage external and internal network configurations to optimize learning outcomes. To optimize the flow of new knowledge, managers need to maintain internal relational cohesion, as well as an external bridging connection outside the learning network. When managing inter-network cohesion, our study also found that increasing the investment commitment with partners is more effective than extending the duration of an alliance.

Second, this study enhances our understanding of the different roles of network embeddedness in alliance networks, both inside and outside innovation ecosystem networks. External bridging ties enhance a focal company’s ability to identify potentially diversified and valuable resources within the entire network. To realize this potential and maintain information flow, companies need to strengthen relational cohesion within their alliance network.

Although our results are encouraging, this study should be interpreted with caution due to some inherent limitations in the patent citation data used. The high level of secrecy and the competitive patent race [73] in the pharmaceutical industry has created difficulties in accessing the latest data for this study. Despite its limitations, the study contributes to our understanding of the complex process of inter-organizational learning by identifying patterns of patent citations that have previously occurred with longitudinal studies that can be generalized. Further research should be undertaken to investigate the variations in network position (such as network centrality, ego betweenness, and structural equivalence) and the relational strength of alliances regarding the scope of collaboration [29]. Future studies may consider these aspects of structural and relational configurations to generate more insight to understand the complexities of managing the network configuration of multiple alliances.

Overall, I conclude that understanding the complex system of network configurations in the innovation ecosystem network requires breaking down the structural and relational configurations. This insight into the complementarity between the two configurations creates opportunities to improve an R&D firm’s position in the network, which, in turn, means that firms can optimize their advantages to be more innovative.