BALANCING INTERNAL AND NETWORK ...

BALANCING INTERNAL AND NETWORK CONSTRAINTS IN ALLIANCE AMBIDEXTERITY DECISIONS

Cristina O. Vlas

Radu E. Vlas

Jindal School of Management

School of Science and Computer Engineering

University of Texas at Dallas

University of Houston-Clear Lake

800 West Campbell Road, SM 43

2700 Bay Area Blvd., D116

Richardson, Texas 75083

Houston, Texas, 77058

[email protected]

[email protected]

BALANCING INTERNAL AND NETWORK CONSTRAINTS IN ALLIANCE AMBIDEXTERITY DECISIONS

Abstract

Previous strategic alliance studies tend to overlook the influence that firm’s routine development can have on the relationship between firm’s position and its propensity to choose certain alliance partners. This study extends ambidexterity research by integrating internal and network perspectives and examines their cumulated effects on firms’ strategic alliance choices. We first acknowledge the interplay between firms’ internal knowledge exploration/exploitation strategy and firms’ alliance formations and find that firms that explore internally are more likely to make focused alliance decisions. Second, our analysis of 145 US-based firms with an active alliance behavior reveals that having well-formed routines as a result of previous collaborations strengthens brokerage firms’ tendency to follow alliances focused on either exploration or exploitation. Although most alliance studies commonly argued in favor of an ambidextrous approach, this study provides critical evidence that both internal knowledge exploration/exploitation strategy and development of routines constrain firms’ alliance formation decisions guiding them towards a more focused approach.

INTRODUCTION What factors fuel firms’ strategic alliance formation decisions? In their decision whether to balance exploratory and exploitative alliances, do firms consider their internal efforts to explore and/or exploit? This question has been tackled by numerous researchers in the areas of strategic management, organizational learning, and knowledge management. The studies that employ the exploration versus exploitation paradigm acknowledge the existence of alternative modes of operation within which firms can achieve balance: internal (through internal knowledge development) or external (through alliances and acquisitions) (Dyer, Kale, & Singh, 2004; Hagedoorn & Wang, 2012; Stettner & Lavie, 2013). Internal balance refers to internal organization of these activities into exploratory versus exploitative learning. Firms build internal knowledge breadth through explorative learning and knowledge depth through exploitative learning (March, 1991). External balance refers to two different ways firms use to either develop and access new knowledge through collaboration (exploration) or commercialize and license their existent products/knowledge (exploitation alliances) (Rothaermel, 2001). More recent research has shown concerns related to balancing exploratory and exploitative strategies at the same time (Lavie, Kang, & Rosenkopf, 2011). Extensive attention has been awarded to separating exploitation from exploration by different means. In a comprehensive literature review of organizational ambidexterity, O'Reilly and Tushman (2013) refer to three different ways in which exploration and exploitation can co-exist: structural, contextual, and sequential. Structural ambidexterity refers to simultaneously involving in exploration and exploitation by using separate units within the same organization (Benner & Tushman, 2003). Contextual ambidexterity refers to achieving balance within the same unit by nurturing adaptability, support, and trust of the individuals (Gibson & Birkinshaw, 2004). Sequential ambidexterity refers to firms’ ability to shift structures over time by adapting their 1

processes (Kauppila, 2010). Brown and Eisenhardt (1997) proposed that exploration and exploitation can be separated temporally. Firms can choose to use a “rhythmic switching” between periods of exploration and periods of exploitation. It has been acknowledged in the sociology and management literatures that firms are relational entities that are embedded in their networks and do not exist in a vacuum, totally separated from other entities (Uzzi, 1996). Their decisions to choose certain alliance partners reflect their past decisions and their embeddedness in their alliance networks (Lin, Yang, & Demirkan, 2007). Prior alliances create a web of relationships among firms, imposing the direction of decisions these firms subsequently make (Granovetter, 1985). Over time, firms might find themselves in the center of the network, thus feeling the constraints of their position. Or, they might find themselves in a position that allows them to facilitate the flow of information between seemingly unconnected actors, position of arbitrage that helps brings them benefits from the intermediation of resources or information (Burt, 1992). We build on these two research views and argue that firms’ decisions to pursue certain alliance partners are influenced by the advantages as well as the constraints resulted from their position in the structure of the network (Ibarra, 1993). Except network embeddedness, internal knowledge development strategy can also have an important effect on firms’ alliance formation decisions. The resources firms have are limited. Firms have to make the call on the best way to spend these resources. Firms that spend their resources to focus on diversifying their knowledge pool will have fewer resources to spend on deepening this knowledge. Firms that choose to spend the resources to specialize in a certain area of interest will have available resources to explore other opportunities. In order to remain competitive on the market, firms will have to find other sources to complement their internal choices. These sources are most often found outside the firm, in their network structure. And 2

then the question is: how do internal and network contexts influence firms’ strategic alliance decisions? One of the most emphasized factors that are probable to affect firms’ ability to balance their internal and external strategic choices brings into the equation firms’ ability to incorporate and apply their knowledge. Adaptation and absorption capacities are both necessary and at the same time exclusive. To adapt better, firms must stay flexible. To absorb better, firms must develop routines. Developing routines limits firms’ ability to adapt but also enhances firms’ ability to exploit. So, inserting business routines into our equation proves that the ambidexterity construct is an even broader and more complex build than we expected. This fact amplifies the complexity of the question this article endeavors to answer: question: how do internal, network, and routine development contexts influence firms’ strategic alliance decisions? We add to the ambidexterity research by studying the influences that firm’s network embeddedness, its internal knowledge development strategy and its routine development have on firm’s intention to explore and exploit via alliance partners. We extend previous research by identifying the most important factors that affect a firm’s alliance formation behavior. To answer our main question, we shape this empirical study as follows. We first present a brief study of the challenges that firms face with regards to aligning and/or adapting their strategies according to their internal and network realities. Second, we explain our framework and incorporate the external and internal constraints that affect firms’ strategic alliance formation decisions. We develop the hypotheses by grounding them into the network, organization, and knowledge management literatures. We find support for some of our hypotheses by testing them on a wideranging dataset covering all alliances formed by 145 focal firms in the communications and information processing industries from 2004 and 2008. We conclude by commenting on the

3

limitations of this paper and future directions for research, and close with our contributions to the field of strategic management. Our overall theoretical framework is illustrated Figure 1. ____________________________ Insert Figure 1 about here ____________________________

THEORETICAL BACKGROUND AND HYPOTHESES DEVELOPMENT

The network context and the moderating effect of business routines There are three important categories of factors that influence a firm’s strategic choice of alliance partners: firms’ network context, their internal knowledge stock, and their ability to learn. A firm will consider both its position in the network (central or arbitrage) and its knowledge capabilities when deciding whether to enter an exploratory partnership, exploitative partnership or pursue both simultaneously. The routines firms developed as a result of previous partnerships will moderate the relationship between firms’ network positioning and their decision to involve in ambidextrous versus focused alliances. The moderating factor that affects firms’ decisions is their capability to integrate routines they currently use or used during previous alliances. Organizational learning and resource-based view researchers acknowledge the fact that firms’ partnerships affect their routine structure. In alliances, firms combine internal and external knowledge (Dyer & Singh, 1998). For the time period of the alliance, firms develop routines that match the nature and needs of that alliance. These routines are stronger when the alliance partners share knowledge in the same business domains and weaker when they don’t. Common business knowledge will help firms better understand each other and will also increase their awareness of the real value and capabilities of 4

their partners. When these alliances end, the routines developed during the partnership will remain as tacit and explicit resources that firms will use in their future alliance decisions.

Firm centrality. Centrality is probably one of the most important network characteristics studied so far (Ahuja, Galletta, & Carley, 2003). Being a central actor has positive as well as negative implications. A high centrality allows the firm to be more connected, to have a better access to resources, to benefit from a high reputation and a higher status. At the same time, being central translated in many relationships that can constrain the firm in its ability to seek new opportunities. For the purpose of capturing the effects of a firm’s central position, we consider both direct and indirect ties between the firm and its partners. Thus, centrality is measured using a closeness-like index. Closeness centrality is the best measure when the researcher wants to take into consideration the number of ties an actor has and also the quality and farness of these partners. Centrality will reflect an actor’s involvement in the cohesiveness of the network. Closeness centrality will also allow us to identify how the diffusion of information inside the network will affect firm’s propensity to engage in ambidextrous alliance formation. Information among actors travels through direct ties but also on indirect ties. Here, we chose to use the average reciprocal distance centrality formulated by Friedkin (1991). This measure defines the distance between two actors as the average length of all the possible paths between them. This takes into account all the ties an actor has to all other actors in the network, whether direct or indirect. We managed to capture the pool of relationships that a central actor can exploit or explore to its advantage. The bigger the pool of connections, the higher the amount of information the central actor has. Firms’ alliance portfolio influences firms’ capabilities to balance internal and external exploration and exploitation of knowledge. Firm’s exploration and exploitation routines further 5

determine firm’s preference for future partner selection. Firms that are strongly embedded in their networks (central firms) find themselves in a position of power, position that allows them to pick and choose their alliance partners. When firms leverage too much on their existent partnerships, they risk facing learning myopia (Levinthal & March, 1993). As long as they can conquer the tendency towards to be around too much within knowledge domain experience which hinders exploration of new knowledge (Perry-Smith & Shalley, 2003), these actors have the position that gives them a good chance at exploring new relationships without hindering performance. Central firms can avoid learning myopia and routines overembeddedness by maintaining a balanced approach in their alliance formation strategy. Central firms have the resources necessary to simultaneously develop and sustain exploitation and exploration routines. A significant part of a central firms’ success resides in their ability to remain flexible and innovative, especially in a high growth market. The effectiveness of exploitation is related to following certain routines that the firm has in place while an ambidextrous approach would imply applying these routines to new knowledge. For example, large firms can handle an ambidextrous approach by using different units within the firm. By maintaining an ambidextrous approach in their alliance formation, these actors will both diversify their routines that would enable them to remain flexible and competitive, and also, leverage their existent routines this to gain efficiency and to secure their market position. Thus, we argue that having well developed routines will strengthen firms’ ambidextrous orientation in their alliance formation decisions. H1: A firm that has a high centrality in the alliance network will benefit from following an ambidextrous approach in the formation of its alliances when it has well developed routines.

6

Structural holes. Firms that play a broker role, mediating the information flow between unrelated actors, find themselves in a temporary position (Burt, 2002). They have to be quick and draw as many advantages as they can from their partners in a very short period of time. They do not have the time to develop routines and they do not have the luxury of abundant resources to help them sustain an ambidextrous alliance approach. Strategic management literature underscored the finding that a focused external approach will prove beneficial for firms rich in structural holes (Lin et al., 2007). To complement existing research, we argue that having developed routines within their knowledge domain (from previous alliances with partners within the same knowledge domain) will strengthen broker firms’ choice of engaging in focused alliances. Timing is of essence for broker firms as they need to choose the right combination of alliance partners that allows them to reap maximum benefits in a short time (usually one year). Forming repeated alliances with different partners either within the same knowledge domain (exploration) or across knowledge domains (exploitation routines) allows brokers to extract the maximum benefits by employing their knowledge routines (either exploitation or exploration) in those domains to extract information quicker. By pursuing both within and across knowledge domains alliances will lead to a low probability to foster consistent practices, possible misapplication of routines, or even negative learning transfer (O’Grady & Lane, 1996). A graphical representation of the above arguments will further help clarify the logic behind the moderating effect of business routines on the relationship between firms’ embeddedness and their approach toward strategic alliance formation. ____________________________ Insert Figure 2 about here ____________________________

7

Figure 2 represents an oversimplification of a possible network structure. Actor A has in its alliance portfolio only firms within the same knowledge domain (SIC 4 – Communication Industry). This actor can develop routines to exploit future alliances within the same domain. Actor B has alliances only firms active in different knowledge domains than itself. This actor can develop exploration routines by accessing information across domains. Actor C has in its alliance portfolio firms both within and across its own knowledge domain. For actor C it takes longer to develop routines because its own knowledge domain is different than its partners’. This actor cannot take the full advantage of control and access to information before the structural holes decays (usually one year). Both actors A and B engaged in repeated alliances yet with different partners but they stayed within their knowledge domains. Actors A and B will benefit if following a focused approach because they have developed routines that help them incorporate new information faster. Actor C engaged in alliances with different partners across different domains and this limits its ability to quickly incorporate the information it was exposed to during these partnerships. Consequently, starting with the assumption the a brokerage position is a very temporary position, we argue that broker firms will follow either an exploration or exploitation focused approach in choosing their partners because this is the only way to reap benefits from the routines they developed. H2: A firm that has a high degree of brokerage positions in the alliance network will benefit from following a focused approach in the formation of its alliances when it has well developed routines.

The internal knowledge orientation In today’s uncertain business environment, innovation is one important dimension essential for firm’s performance. But innovation is a risky business. Researchers highlighted the fact that 8

external sources are extremely important sources of knowledge (Laursen & Salter, 2006; Rosenkopf & Nerkar, 2001). Previous research has shown that some firms will try to stay innovative by expanding their knowledge sources beyond their internal R&D objectives (Baldwin & Clark, 2000) using a “parallel path strategy” (Nelson, 1961, p:351). Very few firms have knowledge concentrated into only one area of activity. Most firms try to stay flexible by concurrently developing knowledge not only in different domains but also in different geographic locations (Ahuja & Katila, 2004; Katila & Ahuja, 2002). They will be open even to knowledge they can retrieve from their customers (Von Hippel, 1986) or suppliers (Leiponen, 2001).

Internal knowledge breadth. The main interest of the second part of this paper is to test whether different internal orientation (exploratory or exploitative) can have different effects on firms’ decision to pursue an ambidextrous versus focused external approach in their search for knowledge sources. Knowledge breadth captures firm’s learning and search across disciplines. Knowledge breadth can be defined as the variety of knowledge a firm has. Increasing products complexity requires a firm to have knowledge in a variety of technological areas (Ernst, 2001). According to March (1991), knowledge breadth can be seen as an appropriate measure for internal exploratory learning. Firms that pursue an exploration strategy will face higher risks due to novelty but it will also have higher rewards from discovering new opportunities. In industries like technology intensive industries (i.e. communication, social media, information retrieval) where firms face high uncertainty, complementing internal exploration with external exploration increases the diversification of firms’ knowledge stock and their chances to be one step ahead of competition at all times. Researchers underscored the importance of external exploration through alliances as 9

a straightforward way to increase the chance of success by broadening the categories of knowledge that a firm pursues (Leiponen & Helfat, 2010). The next question would be: what combination of external sources is most desirable under the constraints imposed by the adoption of a certain internal knowledge strategy? Firms focused on internal diversity (high knowledge breadth) might want to look for external partners that can help deepening their knowledge into some areas. Firms focused on internal sophistication (high knowledge depth) might want to search for partners to help them diversify. At the same time, we have to consider for the fact that firms are rarely at these extremes. Most of them develop knowledge in multiple domains and also further deepen their knowledge in only some domains. We stem from the assumption that knowledge breadth and depth are two separate constructs that are independent of one another. We measure them separately (different scales) and we do not consider them integral parts of the same continuum. Researchers identified that firms’ performance is conditioned by both knowledge breadth and depth (Nelson, 1982). Depth can provide higher performance gains but these gains cannot be sustained without some degree of knowledge breadth. Dosi (1982) argued that performance benefits of a certain piece of knowledge shows evidence of a pattern of decreasing returns as knowledge depreciates. This calls for both depth and breadth of knowledge whether it is built internally or externally. We have a reason to believe that firms are rational actors. In order to learn and remain flexible firms choose to increase their competitive advantage by developing core knowledge in house and complementing their needs by the means of external partnerships (McGrath, 2001). This way firms would save the costs associated with developing the respective capabilities on their own. Similarly, firms may choose to ally with partners for pure exploitative purposes. Each firm would benefit from economies of scale. Depending on their needs, firms have the potential 10

to follow either option. The question is whether they choose to follow these two options simultaneously or sequentially. Consequently, we formulate two competing hypotheses: H3a: A firm with a high level of internal knowledge breadth will tend to follow an ambidextrous approach in the formation of its alliances. H3b: A firm with a high level of internal knowledge breadth will tend to follow a focused approach in the formation of its alliances.

Internal knowledge depth. Whenever firms develop patents in the same class, their knowledge complexity in that specific area of expertise increases. Consequently, firms’ abilities to exploit the knowledge pooled in that area of expertise increases. Following Wang and von Tunzelman (2000, p.806) we understand knowledge depth as the level of “analytical sophistication.” In our view, internal knowledge depth is concerned with the level of sophistication while internal knowledge breadth better reflects the level of firms’ knowledge heterogeneity. Firms that have high knowledge depth also have a good understanding of their domain of activity. They know what works and what doesn’t. Internally, firms develop routines that help them draw benefits from relying on their developed knowledge stock. Externally, we can expect that these firms trust their potential to exploit the knowledge that is shared or spilled over through their partnerships, as long as the knowledge falls under their areas of expertise. They would be inclined to focus on choosing those alliance partners that might offer them the possibility to further exploit the pool of knowledge they already developed within the organization. We define these alliances as focused on exploitation. By choosing exploitation alliances, firms would continue to increase the level of their own knowledge sophistication, to refine their existent knowledge, and to benefit from using the routines they developed over time (Lavie et al., 2011).

11

Firms with high knowledge depth have well established routines in place and probably benefit from the advantage of a good in-house communication and coordination. Being focused in their choice of alliance partners would be more beneficial than being ambidextrous, as firms would pursue a persistent pattern of behavior and benefit most from the effective use of their specialization. Consequently, we would expect that both strategies have potential positive effects on firm’s choice of alliance partners. Following an ambidextrous approach allows firms to avoid obsolescence by tapping into new areas of knowledge while leveraging existent knowledge stock. Following a focused approach encourages firms to either leverage on their exploitation routines by applying and fine-tuning their own knowledge or to develop exploration routines that open the way toward new learning opportunities. As result, we test the following set of competing hypotheses: H4a: A firm with a high level of internal knowledge depth will tend to follow an ambidextrous approach in the formation of its alliances. H4b: A firm with a high level of internal knowledge depth will tend to follow a focused approach in the formation of its alliances.

RESEARCH METHODOLOGY Sample For this study, we identified an industry with active alliances. We selected the US communication and information services industries that are comprised of all wireless and wired telecommunications carriers (SIC 4812 and 4813), data processing and information retrieval services firms (SIC 7372, 7373, 7374, 7375) and business services firms (SIC 7379 and 7389). We chose this diverse data set because it gives us the opportunity to observe firms’ behavior and motives in alliance formation in various high growth industry contexts. Following Todeva and Knoke (2005), in defining our data set we tried to cover all motives for which firms might seek 12

partnerships: learning and competence building (organizational), cost sharing and risk diversification (economic), achieving competitive advantage and gaining access to new technologies (strategic), and developing standards and overcoming regulatory barriers (political). We defined the boundaries of our network following a similar procedure to the one used by Yang, Lin, and Peng (2011): first, we identified all non-equity alliances that occurred among firms with primary SIC code in these industries, second we considered only alliances between US firms that were active in the period between 2004 and 2008; third, among all these alliances, we retained only those alliances that involved at least two members of the selected industries and discarded alliances formed by members of our selected industries only with members of other industries. For example, an alliance between a firm-member of communication or information services industries and 3 partners of other industries was discarded while an alliance that involved at least 2 members of communication or information services industries and also members of other industries was retained. We called these alliances within-industry group alliances. In total, our data set contained 2,219 firms involved in 1,444 alliances with at least one other member of the communication and/or information services industries over a five-year period. On average, firms in our dataset were involved in 1.5 alliances during this period. Only 407 firms were involved in more than 2 alliances during this period and 145 were involved in more than 3 alliances. Considering the fact that testing our hypotheses required an industry rich in the number of alliances formed by each firm and not in the number of firms that are involved on average in only one alliance, defined as exploration by default by Stettner and Lavie (2013), we decided to define our focal firms as firms that were involved in more than three alliances between 2004 and 2008. Therefore, we computed an alliance concentration measure for withinindustry alliances announced between 2004 and 2008. We identified that our 145 focal firms 13

were involved in 547 alliances, roughly a third of the total number of alliances. The aforementioned approach is considered to yield a representative subset of firms with an active and consistent alliance behavior. Firms not included in this subset are firms with one or two alliances over a period of five years of very active alliance formations. These firms might pursue an alliance for a very specific purpose that might occur only once in the entire life of that company, thus being not representative of their strategic behavior. Their alliance formation behavior cannot be generalized to the entire industry and especially to those firms that involve in partnerships as a consistent means to co-evolve with their strategic partners (Koza & Lewin, 1998). The identified focal firms have a steady, habitual alliance behavior targeted to either develop and access new knowledge through collaboration partners (exploration alliances) or market and commercialize products/services based on their existent knowledge (exploitation alliances) (Lavie & Rosenkopf, 2006; Park, Chen, & Gallagher, 2002; Rothaermel, 2001; Rothaermel & Deeds, 2004) . We used SDC Platinum database to retrieve data on alliances as it is one of the most complete databases on US alliances (Schilling & Steensma, 2002). We further verified the alliances formed by our 145 focal firms using LexisNexis. We also used Standard & Poor’s Compustat to collect financial data and WIPO’s PatentScope to collect patent information for each of these firms. The reasoning behind using WIPO’s Patent Scope database and not USPTO or NBER databases is that it retrieves accurate information from USPTO and also offers the possibility to compute aggregated measures by using queries. We designed such computation queries to retrieve aggregated patent information by patent class and in total for all classes of patents each focal firm applied for between 2004 and 2008 inclusive. In order to better capture the cumulative effects of a firm’s alliance portfolio, we use a five-year moving window as suggested by Kogut (1988). The SDC database has relatively 14

complete information on new alliances announced each year since mid-1980s. However, this database doesn’t mention the termination date of these alliances and previous research suggested that a five-year moving window best captures the influences an alliance formed in preceding years might have on firm’s current network embeddedness. For example, the alliance network for 2004 is constructed based on all alliances that firm announced between 2000 and 2004. To be sure that we are capturing the effects of existing alliances on a firm’s decision to follow a focused or an ambidextrous behavior, we also used a one-year lag. As a result, we collected additional alliance data for our focal firms for the period 1999-2003. To compute the network measures, we built a symmetric matrix of 2,219 by 2,219 for each year using Ucinet 6 (Borgatti, Everett, & Freeman, 2002).

Measures Dependent variable Alliance ambidexterity. Our dependent variable is firm’s alliance ambidexterity. In computing this variable, we focused on the type of firm’s previous alliances. We operationalized ambidexterity using a continuous measure for both exploration and exploitation instead of two separate measures. In doing so, we assumed that exploration and exploitation are two separate indicators of activities that inhibit each other (Sidhu, Commandeur, & Volberda, 2007; Simsek, Heavey, Veiga, & Souder, 2009; Uotila, Maula, Keil, & Zahra, 2009). The assumption that both exploration and exploitation activities are conceptualized along a single continuum is also consistent with previous research (Abernathy, 1978; Lavie, Stettner, & Tushman, 2010; March, 1991). A firm cannot pursue a pure exploitation or pure exploration strategy in choosing its alliance partners as it would be detrimental to firm’s performance in the long run. Instead, firms

15

engage in different types of alliances, continuously balancing between exploration and exploitation as their needs require. A firm can form alliances to either explore and gain access to new knowledge or exploit and leverage existent knowledge (Koza & Lewin, 1998; Lavie & Rosenkopf, 2006; Rothaermel & Deeds, 2004). Following Rothaermel and Deeds (2004) we defined each alliance that involved joint R&D activities as exploration, each alliance that involved joint marketing, licensing, resale or production activities as exploitation, and a combination of both activities as half exploration, half exploitation. Exploration alliances have been coded as 1, exploitation alliances as 0 and mixed alliances as 0.5. We then used a moving window of five years and summed up all alliances announced by each firm. We used this coding to compute the exploration index for each focal firm for each year. This index is computed as the ratio of total value of exploration a firm was involved in to the total number of alliances formed in the last five years. For example, a firm that formed five alliances between 2004 and 2008, two exploratory alliances, two exploitative alliances and one mixed alliance will have an exploration index of (1+1+0+0+0.5)/5 = 0.5 for the year 2008. The exploration index can take values between 0 and 1. When this index takes a value between 0.3 and 0.7, then we would consider that a firm formed a balanced number of exploration and exploitation alliances and it would be called that it follows an alliance ambidextrous approach. When this index takes values below 0.3 or above 0.7, then we would say that a firm predominantly follows a focused approach. Lower values of the exploration index (below 0.3) would translate into a strategy focused on exploitation, while higher values of this index (above 0.7) would translate into a strategy focused on exploration. This study intends to compare the implications of internal knowledge management and network embeddedness of firms in communication and information services industries on their 16

ambidextrous versus focused approach on the selection of their alliance partners. Thus, we take the interpretation of firms’ exploration index one step further. Following Lin et al. (2007) we transform this index into a dichotomous variable by coding it as ambidextrous or 1 if the exploration index is between 0.3 and 0.7 and focused or 0 if the index is below 0.3 or above 0.7 (inclusive). Dichotomizing the alliance ambidexterity tells us whether a firm follows an ambidextrous or focused approach in choosing its alliance partners. As we are interested in firm’s alliance formation choices, we assume that these are influenced by current alliance portfolio and thus consider all alliances announced in most recent five years (including current year). Independent variables Business routine. Our main independent variable that also plays the role of moderator is firm’s business routine. The partnerships a firm is involved in leaves a mark on that firm’s routine structure. In alliances, firms combine internal and external knowledge (Dyer & Singh, 1998). When an alliance ends, the routines developed during the partnership will remain as an intrinsic part of the firm, influencing current capabilities and future strategic decisions. Firm’s alliance portfolio influences firm’s capabilities to balance internal and external exploration and exploitation of knowledge. Firm’s exploration and exploitation routines further determine firm’s preference for future partner selection. We used a five year moving window to determine the business routine index for each focal firm. We used the primary SIC code as a proxy for the firms’ main area of specialization that affects the area where firms’ routines are developed. The formula we used measures firms’ degree of involvement in alliances that strengthen their existent routines: 𝑠

[∑𝑡𝑡−4 (𝑞) ] 𝑡⁄ 𝐵𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑟𝑜𝑢𝑡𝑖𝑛𝑒 𝑖𝑛𝑑𝑒𝑥 = 𝑛

(1) 17

We measured whether partners in an alliance are all active in the same business sector (measured by similarity in their primary SIC code). In alliances that involved only two partners, the routine score of both partners would be a 0 (zero) if they have different primary SIC code based on years (t ‒ 4) ‒ (t) moving window and a 1 (one) if they have same SIC code. For alliances that involved more than two partners, we coded each firm business routine score as s/q, where q represents the number of partners in that specific alliance and s represents the number of partners with same SIC code. For example, an alliance involving three different SIC codes yielded an score of 0.33 to each firm while an alliance involving three SIC codes, two of them identical, yielded a 0.66 score for those two firms with identical SICs and a 0.33 score for the firm with a different SIC code. Each firm will have one business routine score for each alliance. To compute the business routine index for each focal firm, we had to consider the entire alliance portfolio of each firm. This portfolio includes all alliances that were announced in previous five years. Each firm’s business routine index is the result from summing up all the scores that firm obtained for each alliance and divided the result by the number of alliances that firm announced in the preceding five years. The result will be a continuous variable that can take values between zero and one. High values of the business routine index translates into a firm being involved mostly in alliances with partners who’s own exploration/exploitation routines will reinforce focal firm’s refinement of specialized resources and routines while low values of this index translates into a firm being involved in alliances with partners that will diversify focal firm’s learning and break from existent routines. Firms with high business routine indices can be seen as interested in exploitation alliances while firms with low business routine indices can be seen as preferring exploration alliances. Centrality. Out of the diverse pool of centrality approaches developed over the years all rooted in Freeman (1979) seminal work, we operationalized centrality with a closeness-like 18

centrality measure. Closeness-like measures of centrality assess the length of path an actor is involved in. We consider that measuring centrality with a closeness-like index best captures both the direct and indirect ties an actor has. At the same time, closeness-like measures of centrality consider not only the number of ties an actor has but also the quality and farness of these partners. In the context of alliance formation, closeness measures of centrality reflect an actor’s involvement in the cohesiveness of the network. We used Friedkin (1991) closeness measure based on immediate effects. A secondary purpose for this study, among other, is to identify the effect that diffusion of information inside the network affects an actor’s propensity to engage in ambidextrous alliance formation. Information among actors travels not only on the shortest paths between actors (direct ties) but also on indirect ties. In computing closeness measure, it makes sense to account for all paths among actors, not only shortest ones. We use the average reciprocal distance (ARD) centrality formulated by Friedkin (1991). This measure defines the distance between two actors as the average length of all the possible paths between them. This takes into account all the ties an actor has to all other actors in the network, whether direct or indirect. For this purpose, we first constructed the non-directional matrix for each year by using a five year moving window (the average age of an alliance is five years (Kogut, 1988)). Second, we used the command for multiple centrality measures in UciNet 6 (Borgatti et al., 2002) to calculate the average reciprocal distance for each focal firm: ∑𝑛 𝑖=1 𝑚𝑖𝑗

𝐴𝑅𝐷𝑗 = (

𝑛−1

−1

)

,𝑖 ≠ 𝑗

(2)

where 𝐴𝑅𝐷𝑗 represents the average reciprocal distance of actor j calculated as the closeness of actor j to all other actors in the network and n is the total number of firms in the network. ∑𝑛𝑖=1 𝑚𝑖𝑗 is the sum of lengths of all possible paths from actor i to actor j in that network. 19

Structural holes. We used Burt (1992) measure of constraint to compute our structural holes variable. Burt’s constraint measure reflects the extent to which focal firms are directly and indirectly connected with other firms in their network. A firm will be constrained by being connected only to firms that are not further connected to other actors in that network. When all ties in a network are concentrated in only one contact, we call the network highly constrained. The formula used to compute our constraint measure is: 𝐶𝑖𝑗 = [𝑝𝑖𝑗 + ∑𝑞(𝑝𝑖𝑞 𝑝𝑞𝑗 )]2 ;

q ≠ i, j

(3)

where 𝑆𝑖𝑗 is the proportion of ties that firm i has with firm j and ∑𝑞(𝑝𝑖𝑞 𝑝𝑞𝑗 ) represents the proportion of other relations that lead firm i back to j, to the extent that the sum across firms q is different than zero. The total in parentheses represents the proportion of ties that are directly or indirectly invested by firm i in its relation with firm j. The constraint measure from Equation (3) takes values from a minimum of (𝑝𝑖𝑗 )2 when firm j is disconnected from all other contacts to a maximum of 1 (one) if firm j is the only contact firm i has. The network constraint index C is summed up across j, ∑(𝑗𝐶𝑖𝑗 ), to capture the lack of structural holes in firm i’s network. We adapted our constraint measure by following Burt’s (2002) finding about bridges’ decay. Our constraint measure is the weighted average of two previous years’ constraint measures. Following Burt’s findings, we considered year (t-1) with a 90% contribution and year (t-2) with a 10% contribution. For example, the constraint measure for year 2008 is the weighted average of the constraint measure of 2007 weighted by 90% and the constraint measure of year 2006 weighted by 10%. The final constraint measure for 2008 is lagged by one year and reflects the decay rate of structural holes by approximately 90% within one year. Following Soda, Usai, and Zaheer (2004) we multiplied the value of constraint by ( ‒1 ) in order to capture the lack of constraint which equates a structural hole. We also multiplied structural holes scores by 100 to facilitate the discussion of the results. It’s important to note that 20

this measure is calculated on the data prior to the alliance announcement event year. By lagging the scores with one year allows us to capture the advantages of structural holes in deciding whether to follow an ambidextrous or focused approach in subsequent alliance formations. Technological knowledge breadth. Technological knowledge breadth (KB) is measured using a continuous variable that ranges from 0 to 1. Breadth is defined as the variety of technological knowledge a firm has. A firm with high knowledge breadth has a wide technological knowledge base. The firm’s knowledge is horizontally spread and it has a wide variety of patents in different classes. Having the knowledge spread over various domains, the firm is considered to have an internal exploration strategy. We use Jose, Nichols, and Stevens (1986) approach and compute technological knowledge breadth and suppose that a firm has its technological knowledge distributed over n classes of patents. The wider the spread of knowledge in various classes, the higher the breadth of technological knowledge that firm has. The formula we used is: 𝐾𝐵 = 1 − 1⁄𝑛

(4)

where n represents the number of classes in which a firm has patents granted. This measure captures firm’s knowledge dispersion and is computed for each firm-year observation in our dataset. Data was retrieved from World Intellectual Property Organization’s service called PatentScope that harbors over 35 million national and international patent documents. Technological knowledge depth. Technological knowledge depth (KD) is measured using a continuous variable defined as the sophistication of knowledge a firm has. A firm with high knowledge depth is highly specialized in only one or very few technology sectors. It has patents concentrated in one or only few classes. Knowledge depth is thus reflected in a higher expertise to develop and integrate firm’s knowledge in that specific area of expertise. We construct our 21

knowledge depth measure following Moorthy and Polley (2010) who adapted Jose et al. (1986) operationalization to also include the spread of patents across patent classes. The formula we used to measure technological knowledge depth reflecting a continuous variable that ranges from 0 to 1is depicted below: 𝐾𝐷 = ∑(𝑝𝑖2 − (1⁄𝑛)2 )

(5)

where 𝑝𝑖 represents firm’s proportion of patents granted in class i, and n represents the total number of classes where the firm has patents granted. We applied the same ranking procedure as in the case of knowledge breadth. Control variables Firm size. We measured firm size using the total number of employees (unit:1,000) reported in Compustat as a proxy. This variable is important because firms with a higher number of employees are more capable to generate, develop, implement, or absorb knowledge and information from their alliance partners. Firms large in size are more likely to benefit from a fruitful collaboration when their partners also have the potential (number of employees) to sustain or develop knowledge. Considering that the focus of this paper is on how a firm’s internal knowledge stock and its network embeddedness affect firm’s capability to involve in both exploration and exploitation at the same time, the number of employees has been lagged by one year. Previous studies showed a high correlation between firm’s total assets, number of employees, and total revenues. In technology intensive high growth industries, not all firms generate positive revenue streams, fact that suggests that using number of employees as proxy for firm size is a reasonable alternative (Shan, Walker, & Kogut, 1994). We believe that whether a robustness check had been performed using alternative measures, it would have yielded consistent results. 22

Firm previous alliance experience. Firm’s alliance experience is an important variable to take into consideration when measuring firm’s propensity to explore and exploit through its partnerships. More alliance experience may enhance firm’s specialization and routines (Haleblian & Finkelstein, 1999; Wang & Zajac, 2007). We measured firm’s previous alliance experience by the number of alliances that firm formed in the preceding five years. Alliance event year. We controlled for time series effect. Although the period of study covered only five years, it is possible that technological shocks affected the industry during this time, effect being an increase in the variations observed in the time effect. Thus, we considered the year an alliance was announced by creating five year dummies. This measure helps us control for unobserved heterogeneity in our panel. Industry. We controlled for inter-industry variation by studying firm’s primary SIC code as defining the industry (communication, information services or business services) in which that firm has its main activity. We created three industry dummies based on the primary SIC code for each focal firm. Firms with primary SIC code 4812 or 4813 were coded “communication,” those with SIC code 7372, 7373, 7374, or 7375 were coded “information,” and those with primary SIC code 7379 or 7389 were coded “business.” Knowledge breadth and knowledge depth dummies. Because the WIPO PatentScope service reports only the first 25 classes of patents for each firm, we had to create dummies to differentiate between firms with patents in 25 classes and those that had patents granted in more than 25 classes. To identify which firms had patents in more than 25 classes, we cross checked with USPTO service and compared the number of patents granted each year. If the number of patents granted (as reported by USPTO) was higher than the number of patents reported by WIPO PatentScope, then we concluded that those patents not reported by WIPO must be in some other classes that are not reported. Similarly, we created dummies for firms that didn’t have 23

patents granted in a certain year and those that reported patents granted in only one class. Firms with patents in one class, n equals 1, will have a knowledge breadth value of 0 and firms with no patents will automatically have knowledge breadth equal 0. To differentiate between these two types of firms, we created dummies. We followed a similar procedure to create dummies for the knowledge depth measure. This method helps us solve two problems: (1) helps us differentiate between firms with patents granted in 25 classes and those with patents granted in more than 25 classes; (2) helps us differentiate between firms with patents in only one class and firms without patents.

ANALYSIS AND RESULTS Analysis This study’s dependent variable is alliance ambidexterity. Since we have a dichotomous dependent variable, we had to run logistic regression analyses. We also have multiple observations for each firm over a period of ten years (including the moving windows) which raises concerns of potential interdependence. To address these concerns, first we lagged all our independent and control variables by one year and second we used the command “xtlogit” in Stata12 package that fits the cross-sectional time-series logistic model (random-effects). Logistic regression is the most appropriate method to predict the outcome of a binary variable.

𝑦𝑖𝑡 =

1 1+𝑒 −(𝛽×𝑋𝑖𝑡+𝜇𝑖 +𝜀𝑖𝑡)

(5)

where 𝑦𝑖𝑡 is the alliance ambidexterity for firm i at time t; 𝑋𝑖𝑡 is a vector of characteristics of firm i at time t. These characteristics include network attributes as firm centrality and structural holes, internal organization attributes as knowledge breadth, knowledge depth,

24

business routines, and general firm characteristics as age, size, and previous alliance experience. 𝜇𝑖 represents a time-invariant effect for firm i which in our model is random. 𝜀𝑖𝑡 is the error term.

Results ____________________________ Insert Table 1 about here ____________________________

Table 1 presents descriptive statistics and the correlations between variables. This table included the number of observations, the mean, standard deviation, minimum, and maximum for each variable. To save space, we did not include the dummies for year, industry and knowledge breadth/depth. In this table there are also presented the correlations between variables. The low level of correlation suggest that our variables are independent. Following Cohen, Cohen, West, and Aiken (2013) we mean-centered the centrality, structural holes and business routine variables before generating their interaction terms. The results obtained after mean-centering were very similar with the results we obtained before mean-centering the variables. This tells us that our variables were behaving and mean-centering is unnecessary. In the final model we decided to use the original variables values. VIF ____________________________ Insert Table 2 about here ____________________________ Table 2 presents the results of hierarchical panel logistic regression on alliance ambidexterity. To test our hypotheses, we build 4 models. We sequentially added variables such that we avoid an increase in multicollinearity. We followed previous research and started by first 25

adding control variables to the model, then predictor variables, then each interaction one at a time (Lin, Yang, & Arya, 2009; Yang et al., 2011). Our Hypotheses expected versus confirmed results are presented in Table 3 below. ____________________________ Insert Table 3 about here ____________________________ In Model 1, we entered in the equation the control variables only: firm age, firm size (as reflected in the number of employees), firm previous alliance experience, alliance year dummies, industry dummies and internal organization (knowledge breadth and depth) dummies. This model shows that firm previous alliance experience has a significant positive effect on firms’ decision to follow an ambidextrous alliance strategy. Firm age also has a significant but negative effect. We can conclude that more mature firms prefer a focused external strategy in choosing their partnerships, by focusing on either exploration or exploitation. These effects, both for firm alliance experience and firm age remain significant over the subsequent three models suggesting that firms’ alliance experience is indeed positively related to alliance ambidexterity while firm age is negatively associated with it. In Model 2, we added the predictor variables: firm centrality, firm structural holes, internal knowledge breadth and depth, and firm business routines. Our first set of competing hypotheses argues that firms with a high level of internal knowledge breadth will tend to follow an external ambidextrous approach (Hypothesis 3a) or a focused approach (Hypothesis 3b). Model 2 shows that engaging simultaneously in both exploration and exploitation alliances has a significant negative effect for firms with a wide internal knowledge breadth (β = ̶ 6.04; p < 0.05), supporting Hypothesis 3b. Our second set of competing hypotheses argues that firms with a high level of internal knowledge depth will tend to follow an external ambidextrous approach 26

(Hypothesis 4a) or a focused approach (Hypothesis 4b). Model 2 does not provide support for any of these (β = ̶ 2.84, p > 0.05). Models 3 and 4 show the results for our interaction terms. Model 3 helps us test Hypothesis 1. Our Hypothesis 1 argues that central firms will benefit from following an ambidextrous approach in the formation of their alliances when they have well developed routines. The results do not show that there is a significant interaction between firm centrality and its business routines. Hypothesis 1 is not supported. Our Hypothesis 2 argues that broker firms will choose an external focused approach, either exploration or exploitation. The interaction between firms’ structural holes and business routines is marginally significant (β = ̶ 0.04, p < 0.1), supporting Hypothesis 2. Figure 3 illustrates the interaction plots. Panel A shows that firms with a high degree of brokerage positions tend to follow a focused approach in the formation of their alliances when they have well developed routines. Panel B shows no interaction effects between firms’ centrality and their routine development. One explanation might be that even if central firms develop routines that allow them to successfully engage in ambidextrous alliance behaviors, there are other, more important factors that these firms consider when they choose their partnerships. On the other side, for firms that play a broker’s role, developing routines is not only a necessity but also a requirement if they want to be able to quickly draw benefits from the short lived advantages offered by being in the arbitrage position. ____________________________ Insert Figure 3 about here ____________________________

DISCUSSION

27

Contributions At least three major contributions emerge. First, this study advances an innovative recent conceptualization of ambidexterity promoted by Stettner and Lavie (2013). This approach view balance across modes of operation (internal organization or alliance mode) as more efficient than balance within each mode of operation. Our finding that firms differ in their alliance formation choices by their internal organization structure and by their network embeddedness supports Stettner and Lavie (2013) work. Second, we offer an unprecedented perspective on the factors that affect alliance ambidexterity. To better understand the decision towards a balanced or focused alliance approach, this study considers three main areas of influences: internal knowledge development, network conjecture, and firms’ routines. Firms do not decide whether to form an exploration or exploitation alliance in complete disregard of their internal capabilities to deal with the new information flow (Burton, Obel, & DeSanctis, 2011; Rivkin & Siggelkow, 2003) or in disregard of their network position (Lin et al., 2007). At the same time, their capabilities to perform well are dependent on the existence of appropriate routines in place. This brings further support to Stettner and Lavie (2013) claim that balance is better to be achieved on different planes of operation and not within. Third, our study advances the understanding of routines as a new moderator factor. We apply the concept of negative learning effect (Novick, 1988; O’Grady & Lane, 1996) to alliance context and argue that firms that developed routines will opt for a focused exploration or exploitation to avoid the misapplication of knowledge especially because these routines help them benefit from filling their partners’ knowledge holes. Limitations and future research

28

This study is not without limitations. First, our targeted industries are high growth technology intensive. The results we observed might not be applicable to more stable industries. Second, when establishing the boundary conditions for our alliance network, we considered only firms with at least three alliances in the five year period covered by our study. Our findings cannot be extended to the behavior of firms that do not have an active alliance strategy. To extend the understanding of ambidexterity construct, future research could further investigate how network ties affect alliance ambidextrous behavior across different modes of operation. Although this study allowed us to compare and observe the interrelations between two different modes of operation (internal and alliance), acquisitions have not been considered. Studying mergers and acquisitions is extremely important as various external factors, such as institutional influences or cultural propensity toward exploration or exploitation, might affect the findings.

CONCLUSION Overall, this study shows that there are three main categories of factors that influence firms’ alliance ambidexterity decisions: internal knowledge orientation, network embeddedness, and firms’ business routines. The proper combination of these three factors contribute to advance our understanding of what are the best conditions when alliance ambidexterity must be pursued. Scholars have debated the best means to achieve balance in alliance formation but they rarely considered the entire spectrum of factors that affect this decision. As a result, the strategic management literature found contrasting results with regards to balancing exploration and exploitation in the alliance mode. This is mainly due to scholars’ disregard of the multitude of factors, internal organization of knowledge, network position and opportunities, and intrinsic firm characteristics and their combinations that might influence a balanced or focused alliance 29

formation strategy. Our study enhances the understanding of these contingencies by highlighting key interactions among some of these factors. Most importantly, this study considers a variety of industries that makes the generalization of results more plausible.

REFERENCES Abernathy, W. J. 1978. The Productivity Dilemma: Roadblock to Innovation in the Automobile Industry. Baltimore: Johns Hopkins University Press. Ahuja, G. & Katila, R. 2004. Where do resources come from? The role of idiosyncratic situations. Strategic Management Journal, 25(8‐ 9): 887-907. Ahuja, M. K., Galletta, D. F., & Carley, K. M. 2003. Individual centrality and performance in virtual R&D groups: An empirical study. Management Science, 49(1): 21-38. Baldwin, C. Y. & Clark, K. B. 2000. Design rules: The power of modularity. Cambridge, MA: MIT Press. Benner, M. J. & Tushman, M. L. 2003. Exploitation, exploration, and process management: The productivity dilemma revisited. Academy of Management Review, 28(2): 238-256. Borgatti, S. P., Everett, M. G., & Freeman, L. C. 2002. Ucinet for Windows: Software for social network analysis. Harvard, MA: Analytic Technologies. Brown, S. L. & Eisenhardt, K. M. 1997. The art of continuous change: Linking complexity theory and time-paced evolution in relentlessly shifting organizations. Administrative Science Quarterly, 42(1): 1-34. Burt, R. 1992. Structural holes: The social structure of competition. Cambridge, MA: Harvard University Press. Burt, R. S. 2002. Bridge decay. Social Networks, 24(4): 333-363. Burton, R. M., Obel, B., & DeSanctis, G. 2011. Organizational Design: A Step-by-step Approach. Cambrigde, MA: Cambridge University Press. Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. 2013. Applied Multiple Regression/Correlation Analysis for the Behavioral Sciences. Routledge, NY: Taylor & Francis. Dosi, G. 1982. Technological paradigms and technological trajectories: a suggested interpretation of the determinants and directions of technical change. Research Policy, 11(3): 147-162. Dyer, J. H. & Singh, H. 1998. The relational view: Cooperative strategy and sources of interorganizational competitive advantage. Academy of Management Review, 23(4): 660-679. Dyer, J. H., Kale, P., & Singh, H. 2004. When to ally and when to acquire. Harvard Business Review, 82(7-8): 109-115. Ernst, H. 2001. Patent applications and subsequent changes of performance: Evidence from timeseries cross-section analyses on the firm level. Research Policy, 30(1): 143-157. Freeman, L. C. 1979. Centrality in social networks: Conceptual clarification. Social Networks, 1(3): 215-239. Friedkin, N. E. 1991. Theoretical foundations for centrality measures. American Journal of Sociology, 96: 1478-1504. 30

Gibson, C. B. & Birkinshaw, J. 2004. The antecedents, consequences, and mediating role of organizational ambidexterity. Academy of Management Journal, 47(2): 209-226. Granovetter, M. 1985. Economic action and social structure: The problem of embeddedness. American Journal of Sociology, 91: 481-510. Hagedoorn, J. & Wang, N. 2012. Is there complementarity or substitutability between internal and external R&D strategies? Research Policy, 41(6): 1072-1083. Haleblian, J. & Finkelstein, S. 1999. The influence of organizational acquisition experience on acquisition performance: A behavioral learning perspective. Administrative Science Quarterly, 44(1): 29-56. Ibarra, H. 1993. Network centrality, power, and innovation involvement: Determinants of technical and administrative roles. Academy of Management Journal, 36(3): 471-501. Jose, M. L., Nichols, L. M., & Stevens, J. L. 1986. Contributions of diversification, promotion, and R&D to the value of multiproduct firms: A Tobin's q approach. Financial Management, 15(4): 33-42. Katila, R. & Ahuja, G. 2002. Something old, something new: A longitudinal study of search behavior and new product introduction. Academy of Management Journal, 45(6): 11831194. Kauppila, O.-P. 2010. Creating ambidexterity by integrating and balancing structurally separate interorganizational partnerships. Strategic Organization, 8(4): 283-312. Kogut, B. 1988. Joint ventures: Theoretical and empirical perspectives. Strategic Management Journal, 9(4): 319-332. Koza, M. P. & Lewin, A. Y. 1998. The co-evolution of strategic alliances. Organization Science, 9(3): 255-264. Laursen, K. & Salter, A. 2006. Open for innovation: The role of openness in explaining innovation performance among UK manufacturing firms. Strategic Management Journal, 27(2): 131-150. Lavie, D. & Rosenkopf, L. 2006. Balancing exploration and exploitation in alliance formation. Academy of Management Journal, 49(4): 797-818. Lavie, D., Stettner, U., & Tushman, M. L. 2010. Exploration and exploitation within and across organizations. Academy of Management Annals, 4(1): 109-155. Lavie, D., Kang, J., & Rosenkopf, L. 2011. Balance within and across domains: The performance implications of exploration and exploitation in alliances. Organization Science, 22(6): 1517-1538. Leiponen, A. 2001. Why do firms not collaborate? The role of competencies and technological regimes. In A. Kleinknecht & P. Mohnen (Eds.), Innovation and Firm Performance: Econometric Exploration of Survey Data: 253-277. London, UK: Palgrave. Leiponen, A. & Helfat, C. E. 2010. Innovation objectives, knowledge sources, and the benefits of breadth. Strategic Management Journal, 31(2): 224-236. Levinthal, D. A. & March, J. G. 1993. The myopia of learning. Strategic Management Journal, 14(2): 95-112. Lin, Z., Yang, H., & Demirkan, I. 2007. The performance consequences of ambidexterity in strategic alliance formations: Empirical investigation and computational theorizing. Management Science, 53(10): 1645-1658. Lin, Z. J., Yang, H., & Arya, B. 2009. Alliance partners and firm performance: Resource complementarity and status association. Strategic Management Journal, 30(9): 921-940. March, J. G. 1991. Exploration and exploitation in organizational learning. Organization Science, 2(1): 71-87. 31

McGrath, R. G. 2001. Exploratory learning, innovative capacity, and managerial oversight. Academy of Management Journal, 44(1): 118-131. Moorthy, S. & Polley, D. E. 2010. Technological knowledge breadth and depth: Performance impacts. Journal of Knowledge Management, 14(3): 359-377. Nelson, R. R. 1961. Uncertainty, learning, and the economics of parallel research and development efforts. Review of Economics and Statistics, 43(4): 351-364. Nelson, R. R. 1982. The role of knowledge in R&D efficiency. Quarterly Journal of Economics, 97(3): 453-470. Novick, L. R. 1988. Analogical transfer, problem similarity, and expertise. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14(3): 510. O'Reilly, C. A. & Tushman, M. L. 2013. Organizational ambidexterity: Past, present, and future. Academy of Management Perspectives, 27(4): 324-338. O’Grady, S. & Lane, H. W. 1996. The psychic distance paradox. Journal of International Business Studies, 27(2): 309-333. Park, S. H., Chen, R. R., & Gallagher, S. 2002. Firm resources as moderators of the relationship between market growth and strategic alliances in semiconductor start-ups. Academy of Management Journal, 45(3): 527-545. Perry-Smith, J. E. & Shalley, C. E. 2003. The social side of creativity: A static and dynamic social network perspective. Academy of Management Review, 28(1): 89-106. Rivkin, J. W. & Siggelkow, N. 2003. Balancing search and stability: Interdependencies among elements of organizational design. Management Science, 49(3): 290-311. Rosenkopf, L. & Nerkar, A. 2001. Beyond local search: boundary‐ spanning, exploration, and impact in the optical disk industry. Strategic Management Journal, 22(4): 287-306. Rothaermel, F. T. 2001. Incumbent's advantage through exploiting complementary assets via interfirm cooperation. Strategic Management Journal, 22(6‐ 7): 687-699. Rothaermel, F. T. & Deeds, D. L. 2004. Exploration and exploitation alliances in biotechnology: A system of new product development. Strategic Management Journal, 25(3): 201-221. Schilling, M. A. & Steensma, H. K. 2002. Disentangling the theories of firm boundaries: A path model and empirical test. Organization Science, 13(4): 387-401. Shan, W., Walker, G., & Kogut, B. 1994. Interfirm cooperation and startup innovation in the biotechnology industry. Strategic Management Journal, 15(5): 387-394. Sidhu, J. S., Commandeur, H. R., & Volberda, H. W. 2007. The multifaceted nature of exploration and exploitation: Value of supply, demand, and spatial search for innovation. Organization Science, 18(1): 20-38. Simsek, Z., Heavey, C., Veiga, J. F., & Souder, D. 2009. A typology for aligning organizational ambidexterity's conceptualizations, antecedents, and outcomes. Journal of Management Studies, 46(5): 864-894. Soda, G., Usai, A., & Zaheer, A. 2004. Network memory: The influence of past and current networks on performance. Academy of Management Journal, 47(6): 893-906. Stettner, U. & Lavie, D. 2013. Ambidexterity under scrutiny: Exploration and exploitation via internal organization, alliances, and acquisitions, Paper accepted in Strategic Management Journal, and published online. 10.1002/smj.2195. Todeva, E. & Knoke, D. 2005. Strategic alliances and models of collaboration. Management Decision, 43(1): 123-148. Uotila, J., Maula, M., Keil, T., & Zahra, S. A. 2009. Exploration, exploitation, and financial performance: Analysis of S&P 500 corporations. Strategic Management Journal, 30(2): 221-231. 32

Uzzi, B. 1996. The sources and consequences of embeddedness for the economic performance of organizations: The network effect. American Sociological Review, 61: 674-698. Von Hippel, E. 1986. Lead users: a source of novel product concepts. Management Science, 32(7): 791-805. Wang, L. & Zajac, E. J. 2007. Alliance or acquisition? A dyadic perspective on interfirm resource combinations. Strategic Management Journal, 28(13): 1291-1317. Wang, Q. & von Tunzelmann, N. 2000. Complexity and the functions of the firm: Breadth and depth. Research Policy, 29(7): 805-818. Yang, H., Lin, Z. J., & Peng, M. W. 2011. Behind acquisitions of alliance partners: Exploratory learning and network embeddedness. Academy of Management Journal, 54(5): 10691080.

33

APPENDIX

Figure 1. Theoretical framework

Business routine

H1

Firm closeness centrality

H2

Firm structural holes Knowledge breadth

Alliance ambidexterity H3a/b

H4a/b

Knowledge depth

Note: Dashed arrow denotes a relationship that was not tested.

34

Figure 2. Business routine development: within versus across knowledge domains

A SIC 7

SIC 4 C

B

35

Figure 3. Theoretical framework with results

Business routine

0.00 (1.35)

Firm closeness centrality

-0.04 (-1.86 Ϯ)

Firm structural holes Knowledge breadth

Alliance ambidexterity -6.04 (-2.27*)

-2.84 (-0.80)

Knowledge depth

Note: -

Dashed arrow: not tested Regular arrow: hypothesis not supported Bold arrow: hypothesis supported

36

Table 1. Descriptive statistics and correlations for the focal firms 2004-2008 1.

Variables Ambidexterity

2. 3. 4.

N

Mean

SD

Min

Max

583

0.61

0.48

0

1

Centrality

725

0.18

0.52

0

3.61

Structural Holes

725

-27.89

42.20

-122.5

0

-0.01

-0.01

0.96

0.01

-0.03

-0.05

0.78

0.01

-0.04

0.01

0.51***

1

0.06

0.02

0.01

-0.02

5.73

0.04

0.06

- 0.15*

0.07

3.71

0.24***

- 0.01

- 0.12**

0.31***

5.15

-0.07

- 0.01

0.15**

Knowledge 498 0.66 0.31 0 breadth 5. Knowledge 498 0.27 0.16 0 depth 6. Business 583 0.51 0.38 0 routines 7. Firm size 277 0.72 2.21 -6.91 8. Firm alliance 583 1.07 0.58 0 experience 9. Firm age 708 2.46 0.89 0 Significance levels: *p < 0.05; **p < 0.01; ***p < 0.001

1

2

3

4

5

6

7

8

0.10*

0.01

0.16** - 0.09 0.15** - 0.02

-0.20** 0.04

0.30***

-0.07

0.41***

0.19***

37

Table 2. Hierarchical logistic regression model (random effects)a Variables Control variables Firm size Firm alliance experience Firm age Predictor variables Centrality Structural holes Internal knowledge breadth Internal knowledge depth Business routines Interactions Business routines × Centrality Business routines × Structural holes N Wald χ2 Log-likelihood

Model 1 β z 0.03 2.21 -1.36

245 26.91 -122.09

-0.17 3.96 *** -2.27 *

Model 2 β z

Model 3 β z

Model 4 β z

-0.07 4.59 -1.92

-0.31 4.05*** -2.13*

-0.07 4.47 -1.86

-0.30 4.05 *** -2.13 *

-0.06 4.68 -1.78

-0.28 4.08 *** -2.00 *

0.00 0.00 -6.04 -2.84 1.82

2.11* 0.00 -2.27* -0.80 1.48

0.00 0.00 -5.83 -2.79 1.19

0.34 0.17 -2.25 * -0.80 0.94

0.00 0.02 -6.11 -3.22 0.82

1.91 Ϯ 1.47 -2.27 * -0.89 0.64

0.00

1.35 -0.04 214 23.77 -87.92

-1.86 Ϯ

214 23.77 -87.92

214 23.27 -88.68

a

The dependent variable is alliance ambidexterity measured at the focal firm level. Year dummies, industry dummies, and knowledge breadth and depth dummies are included, but not reported here. Ϯ p < 0.1;*p < 0.05; **p < 0.01; ***p < 0.001

38

Table 3. Summary of hypotheses and empirical conclusions Hypothesis H1: A firm that has a high centrality in the alliance network will benefit from following an ambidextrous approach in the formation of its alliances when it has well developed routines. H2: A firm that has a high degree of brokerage positions in the alliance network will benefit from following a focused approach in the formation of its alliances when it has well developed routines. H3a: A firm with a high level of internal knowledge breadth will tend to follow an ambidextrous approach in the formation of its alliances. H3b: A firm with a high level of internal knowledge breadth will tend to follow a focused approach in the formation of its alliances. H4a: A firm with a high level of internal knowledge depth will tend to follow an ambidextrous approach in the formation of its alliances. H4b: A firm with a high level of internal knowledge depth will tend to follow a focused approach in the formation of its alliances.

Expected sign +

Empirical conclusions Not supported

_

Supported

+

Not supported

_

Supported

+

Not supported

_

Not supported

39

Figure 4. Interaction Effects of Business Routines Development Stage and Network Embeddedness Panel A: Firm Business Routines × Firm Structural Holes

Ambidexterity

Underdeveloped business routines Well developed business routines

Low SH

High SH

Ambidexterity

Panel B: Firm Business Routines × Firm centrality

Underdeveloped business routines Well developed business routines

Low Centrality

High Centrality 40