MASTER THESIS The Antecedents of Organizational Ambidexterity: An Examination of Their Synergies and Incompatibilities. Abstract

MASTER THESIS

The Antecedents of Organizational Ambidexterity: An Examination of Their Synergies and

Incompatibilities.

Abstract

It is widely argued that organizational ambidexterity is important to survive or even flourish now and in the future. However, the ambidexterity literature did not reach consensus yet on how to manage organizational ambidexterity. Following the resource-based view, this study aims at solving this dilemma by revealing synergies and incompatibilities among ambidexterity antecedents. Three meta-analyses have been used as starting point, followed up by a path analysis. The path analysis confirmed that synergies and incompatibilities among variables indeed do exist. When combined with all antecedents, market orientation, slack resources and formalization turned out to be related positively to exploitation and exploration. However, other antecedents, such as learning and connectedness, when combined, became significantly negative to exploitation and/or exploration, while alone they were related significantly positive. Hence, this study also contributes to the organizational learning theory, because contrary to the expectations, learning turned out to be detrimental for exploitation when combined with the other antecedents. However, this study did confirm that organizational learning is indeed the driving factor behind exploration and the flexibility of an organization to adapt to environmental changes. Furthermore, an optimization method helps managers to interpret these findings by offering a road map to ambidexterity tailored to their current situation. This study should serve as a building block for future research aimed at solving the ambidexterity dilemma: future research should more comprehensively and accurately grasp the antecedents in the ambidexterity literature to reveal other potential synergies and incompatibilities.

MSc BA Strategic Innovation Management

University of Groningen Faculty of Economics and Business

June 2016 Igor Dolfing S2797585 Word count: 15.220

TABLE OF CONTENTS

1. Introduction……….. 3

2. Theoretical background……… 5

2.1 Ambidexterity from an organizational learning perspective………..5

2.2 Ambidexterity from a resource-based perspective……….6

2.3 Complementarities between organizational learning and the resource-based perspective… 7

2.4 Paradoxical tensions of ambidexterity: exploitation and exploration……… 8

3. Hypotheses development……….. 9

3.1 Learning………. 9

3.2 Market orientation……….. 10

3.3 Resources………... 11

3.4 Coordination mechanisms……….. 11

3.5 Synthesis: potential synergies or incompatibilities……… 14

4. Methodology……….14

4.1 Data collection………14

4.2 Sample……… 17

4.3 Measurements……… 17

4.4 Data analysis methods……… 20

5. Results……….. 23

6. Discussion……… 27

6.1 Managerial implications………. 30

7. Limitations and future research……… 30

8. Conclusion……… 32

References……… 33

Appendix A: Overview results meta-analyses………..43

Appendix B: Overview definitions constructs of interest……… 45

Appendix C: Overview selected studies………... 46

Appendix D: LISREL Program Path Analysis………. 48

Appendix E: MATLAB Program Multivariate Conditional Distribution……… 50

Appendix F: MATLAB Program Quadratic Programming………..52

1.

INTRODUCTION

How do organizations survive and even flourish in the face of constantly shifting market conditions? For an organization to be able to continuously adapt to these conditions, a company should manage the right balance between exploitative processes and explorative processes (Birkinshaw & Gibson, 2004). Exploitation (ET) refers to incremental innovation fostered by processes such as the implementation, and the use and refinement of existing knowledge and skills (March, 1991). These processes in turn intensify the efficiency of the established knowledge and competencies (Zhou & Wu, 2010). Exploration (ER) refers to radical innovation, enabled by processes such as the pursuit of substantively new knowledge and skills, for which experimentation, flexibility and risk taking are necessary (Danneels, 2002; Jansen, Van Den Bosch & Volberda, 2006; March, 1991). The ability of an organization to simultaneously pursue both ET and ER is defined as organizational ambidexterity (OA) (O'Reilly & Tushman, 2004). The resource-based view (RBV) is a prominent theory in the OA literature that suits well to gain a better understanding of OA (Venkatraman, Lee & Iyer, 2007). According to the RBV, due to limited resources firms have to face the tension between exploiting what they know and exploring what they do not know yet, because both ET and ER are critical for their long-term survival (Benner & Tushman, 2003; Chen, Li & Evans, 2012; March, 1991). ET facilitates short-term growth, and allows organizations to survive at this very moment. Furthermore, ET allows organizations to invest in more risky and long-term projects, such as ER, which are necessary for long-term growth. Another influential theory to look at OA is the organizational learning theory (OL) (March, 1991). According to OL, learning capabilities form the engine behind both ET and ER. For organizations it is critical to learn how to adapt in the face of change to stay profitable now and in the future, and therefore, enable a firm to sustain its competitive advantage (Sirmon, Ireland & Hitt, 2007). And indeed, in line with these theories, Junni, Sarala, Taras and Tarba (2013) provide evidence for the OA-performance relationship.

have together a stronger effect on ET and/or ET than they would have had being applied separately. This phenomenon refers to the rise of synergies. But combined resources not only yield potential synergies, they could also constrain each other, which refers to incompatibilities between resources. Therefore, I argue that in order to find how OA can be managed, a more holistic approach is required. The purpose of this study is to combine the myriad of studies that explore antecedents of ET and/or ER, and subsequently, reveal potential synergies and incompatibilities among the antecedents to be able to confirm which antecedents are more important than others to attain this balance.

In order to develop a holistic view on how OA can be managed, this study should not only include the relations between different antecedents and their relation with ET and ER. It should also include the effects these antecedents have on each other. Therefore, it is first of all important to complete a correlation matrix, which reveals not only the correlations between antecedents and respectively ET and ER, but should also be able to reveal the impact these antecedents have on the effectiveness of other antecedents in influencing ET and ER. To determine the antecedents this study should take into account, this study built further on three prior meta-analyses that attempted to capture all relevant antecedents of ET and ER in the literature field (De Boer, 2014; Gubler, 2012; Telman, 2012). Because these meta-analyses do not reveal necessarily all relations the antecedents have among each other, additional papers have been collected to be able to complete the correlation matrix. To be able to reveal the respective importance of the antecedents of ET and ER, their potential synergies and/or incompatibilities, a model had to be tested with two dependent variables (ET and ER) and multiple independent variables. To calculate the relative importance of the antecedents and their potential synergies, both the direct and indirect effects these antecedents have on ET and ER had to be taken into account. More specifically, the direct effect of the independent variables on ET (ER) should be measured, but also their indirect effect on ET (ER), via ER (ET). Therefore, the model included a link between ET and ER. A path analysis has been conducted because it is the appropriate tool to take these direct and indirect effects into account, and showed that there are indeed several synergies and incompatibilities that were not yet revealed in literature before. Additionally, to increase the practical value of this research, I developed an optimization method which is able to work with the input of an organization's current values of ET and ER, and its current values of the antecedents. For example, if an organization currently emphasizes ET over ER (e.g. uses 75% ET and 25% ER) and this organization measures to what extent they have the antecedents of ET and ER in place (the current values of the independent variables), then the optimization method reveals how the antecedents of ET and ER have to be adjusted in order to reach the desired value of ET and ER (e.g.: 50% ET and 50% ER). Thereby organizations know how to maintain or improve the balance between ET and ER to gain and sustain competitive advantage.

this study confirms RBV’s contention that when multiple resources are taken together, synergies and incompatibilities arise between multiple resources and ET and ER. In fact, some of the antecedents that had positive relationships with ET and ER individually, turned out to be negative predictors when combined, and vice versa. Therefore, it shows more comprehensively how to influence ET and ER, and in turn, how to gain and sustain a competitive advantage. Third, this study contributes to OL. This study increases the knowledge on the set of organizational antecedents that facilitate learning processes necessary to foster ET and ER. And confirms that indeed, organizational learning is a driving factor behind ER, and thus, the organization’s flexibility necessary in the face of change, in order to sustain its competitive advantage on the long-term. However, this study also finds that learning in relation to ET, is more complicated than previously thought.

This study consists of five sections. In the next section I situate this study in the existing organizational ambidexterity literature. Furthermore, I will more comprehensively explain the link between OA, the RBV and OL. In the third section I discuss the methodology. Subsequently, in the fourth section the results are presented. Finally, in the fifth section, I discuss the theoretical and managerial implications and the limitations of this study, accompanied with interesting recommendations for future research.

2.

THEORETICAL BACKGROUND

2.1 Ambidexterity from an organizational learning perspective

If some organizations are able to survive in the face of change, and others do not, then it is important to know how some organizations adapt to these changing conditions. A firm’s ability to exploit their current resources to gain revenue and profit, while concurrently exploring new technologies, skills and markets, is essential to a firm’s ability to adapt to changing conditions (Helfat & Peteraf, 2003; Holmqvist, 2004; March, 1991). Duncan (1976) was the first that coined this as organizational ambidexterity (OA), and March (1991) was the first that referred to this as the ability to manage both exploitative innovation (ET) and exploratory innovation (ER). ET, analogous to incremental innovation, refers to improved existing processes and products in existing markets, and ER, analogous to radical innovation, refers to new products and processes for new market domains (Jansen et al., 2006; Liu & Xie, 2014).

(Koçoğlu, Imamoğlu & Ince, 2012). Nevertheless, how these theories are exactly related requires more extensive empirical research (Bustinza, Molina & Arias-Aranda, 2010).

Managing both ET and ER has been proposed as one of the main themes in the organizational learning theory (Levinthal & March, 1993). One of the reasons for this is that innovation only exists with the creation of new knowledge, which is a part of the organizational learning process (Argote, 2012). From the organizational learning perspective, ET can be viewed as the result of exploitative learning activities that symbolize the ability to benefit from existing resources, and allocating them to improve the existing products and processes within existing markets (March, 1991). It also includes refining existing resources to develop competitive advantages, which is in line with the RBV (Barney, 1991; Grant, 1991, 1996). The returns of ET are expected to be predictable and short-term (March, 1991). ER can be regarded as the result of exploratory learning activities that symbolize the ability to discover and try out new ways of doing things and includes things as searching, experimentation, variation, risk-taking and flexibility (March, 1991). ER is related to opportunities of development beyond organizational boundaries, and thus, is interconnected with the environment in which the firm seeks to pursue new knowledge and resources (Holmqvist, 2004; Lavie & Rosenkopf, 2006). This includes the network in which the firm is active, the established alliances of the firm and their potential interorganizational synergies (Lavie, Kang & Rosenkopf, 2011). Consequently, the capability to explore can be seen as the main capability that enables firms to be flexible and reconfigure its existing resource base. The returns of ER are expected to be variable and long-term (March, 1991).

Exploitative and explorative learning capabilities by itself do not directly influence innovative performance, however, they are considered as the engine behind the exchange, use and combination of knowledge and resources into products that are valuable for customers (Atuahene-Gima & Murray, 2007; Holmqvist, 2004; Wei, Yi & Guo, 2014). Or in other words, they provide the necessary incremental and radical innovations, and consequently, generate financial performance advantages (Porter, 1991). How and why innovation can lead to performance advantages will be outlined next argued from the resource-based perspective.

2.2 Ambidexterity from a resource-based perspective

The resource-based view (RBV) argues that the performance differences among firms result from heterogeneity and immobility of the resources, that subsequently can be used to create distinctive capabilities. This applies to resources that are specific to the firm, perceived as valuable to customers, and should have the benefit of being difficult to imitate and non-substitutable (VRIN) (Barney, 1991). Important is the notion by Sirmon and Hitt (2003), that in order to gain and sustain a competitive advantage, the resource inventory in itself is insufficient and often not VRIN. In order to sustain a competitive advantage, first, resources should be bundled into capabilities, and in turn, these capabilities should be effectively leveraged to seize market opportunities (Rugman & Verbeke, 2002; Sirmon et al., 2007). With this notion they extend the RBV and refer to it as the resource management perspective. To explain why this addition is of critical importance, the firm's environment should be taken into account, something the RBV not sufficiently acknowledges (Sirmon et al., 2007).

and bundle their resources into capabilities that allow for seizure of market opportunities. This enables firms to seize new market opportunities constantly, which generates one temporary competitive advantage after another (Sirmon et al., 2007). This way they can create new value effectively, while maintaining the value already created efficiently. Therefore, it is the effectiveness and efficiency by which capabilities are managed that yields and sustains a competitive advantage (Ireland & Webb, 2006).

Creating new value effectively is analogous with the concept of ER, and efficiently maintaining and refining the value already created corresponds with the concept of ET (Levinthal & March, 1993). To create value for customers on the short-term and long-term, incremental and radical innovation is required, firms have to possess both exploitative and explorative capabilities (Levinthal and March, 1993; March, 1991). Following the RBV, OA can be thought of as a distinctive capability with strategic potential, because it allows the firm to exploit today’s resources to create superior value for current expressed needs of customers, while it simultaneously allows the firm to explore how to create tomorrow’s resources to provide superior value for the latent needs of customers. It is argued that firms that excel at managing both ET and ER, are more likely to sustain a competitive advantage (Gupta & Malhotra, 2013). And in line with theory, evidence shows that OA, mediated via ET and ER, influences firm performance, survival rates and innovation rates (Junni et al., 2013; O’Reilly & Tushman, 2013). This is especially the case under the condition of an uncertain market (Caspin-Wagner, Ellis & Tishler, 2012; Goosen, Bazzazian & Phelps, 2012; Jansen et al., 2006), if the concerning firm has sufficient resources (Goosen et al., 2012; Tempelaar & Van de Vrande, 2012) and for larger firms (Yu & Khessina, 2012; Lin, Yang & Demirkan, 2007).

2.3 Complementarities between organizational learning and the resource-based perspective

These theories together demonstrate more comprehensively how and why OA as distinctive capability can lead to sustained superior performance. The RBV (Barney, 1991) explains why distinctive capabilities are valuable and able to sustain a competitive advantage. However, it does not explain how these distinctive capabilities are being developed. Additionally, the RBV depicts a too static picture of the reality, because in fact, fast-changing market conditions are very likely. The resource management perspective extends the RBV and does take the development of capabilities into account (Sirmon et al., 2007). They explain that firms within an uncertain environment have to reconfigure their resource base regularly, and in turn, develop new capabilities necessary to leverage new market opportunities. Subsequently, OL explains how to use and combine resources, therefore it represents the engine behind the development of capabilities and how these capabilities can be adjusted over time to changing environmental conditions (Andreu & Ciborra, 1996; Sirmon et al., 2007).

is referred to as its path-dependency (Barney, 1991). Therefore, organizational learning complements the static orientation of the RBV, and helps to explain how to sustain a competitive advantage in the face of fast-changing markets (Andreu & Ciborra, 1996).

2.4 Paradoxical tensions of ambidexterity: exploitation and exploration.

It is widely argued that managing OA is challenging (e.g.: Junni et al., 2013; March, 1991; Simsek et al., 2009; Tushman & O’Reilly, 1997). First, because it requires management to regulate completely different and inconsistent organizational processes underlying ET and ER, and second, because these processes are reinforcing themselves. The processes underlying ET are focused on improving what has currently been done. Over time, a firm increases its experience in its current capabilities, and in turn, it increases the efficiency of these capabilities. The more an organization is successfully exploiting its current capabilities, the more likely a firm is certain of its short-term returns. Consequently, a firm is more likely to develop organizational inertia. This means firms are less likely to switch to something new, because current success causes firms to overlook (Nelson & Winter, 1982) or disregard (Hannan & Freeman, 1984) new knowledge that is beyond its current trajectory (Zhou & Wu, 2010). Firms overlook simply because they cannot imagine how their current competitive advantage will not endure in the future. And firms might disregard new knowledge, because it takes risk to switch from its current efficient capabilities to new capabilities, since it is not entirely sure that firms will reach the same or higher performance by switching. And since people are in general risk-averse in nature, firms rather strengthen an established position, which thus far has proven well (Zhou & Wu, 2010), and often even tend to overuse ET (Uotila, Maula, Keil & Zahra, 2009). Therefore, ET reinforces ET (Lavie & Rosenkopf, 2006; March, 1991). The over-use of ET is also referred to as the “success trap” (Levinthal & March, 1993), because it lowers the organizations’ flexibility to adapt to changing market conditions, and hence, causes poor performance in the long run (Smith & Tushman, 2005; Levinthal & March, 1993).

On the other hand, the processes underlying ER promote learning from outside experiences. And with successful ER, what currently has been done could be entirely replaced with totally new and more effective capabilities. Therefore, ER constrains the ability to further enhance the efficiency of existing capabilities (Lavie & Rosenkopf, 2006; March, 1991). However, an abundance of ER replaces recent innovations with new ideas, before they had the chance to realise sufficient revenue. This is referred to as the “failure trap” (Levinthal & March, 1993). Because ER has a reinforcing effect on itself as well, it is not unthinkable to fall into this trap. This is because once an organization gains experience in looking outside firm boundaries for new knowledge, a firm becomes better at gathering external knowledge. And once a firm’s skill to acquire external knowledge increases, the number of opportunities to gain new external knowledge increases as well, and with it a firm’s likelihood to do so (Lavie & Rosenthal, 2006). In line with theory, evidence shows that both a lack of or an abundance in either ET or ER, comes at a cost (O’Reilly & Tushman, 2013). Although both ET and ER are critical for survival and wealth, due to limited available resources firms are urged to favour ET over ER or vice versa (Lavie, Stettner & Tushman, 2010). Even though an imbalance between ET and ER can harm the long-term performance of an organization (Levinthal & March, 1993).

example via alliances, allows firms to avoid the inefficiencies that emerge from trying to combine these forces within organizational boundaries (Lavie & Rosenkopf, 2006). However, for the same reason, ambidexterity could be managed within the firm, by dividing ET and ER over different departments. Separating ET and ER within the organization in different departments, is defined as structural ambidexterity (Gibson & Birkinshaw, 2004). Some scholars argue ambidexterity could even be managed within teams, where individuals tend to either exploit or explore (Smith & Tushman, 2005). This does not diminish the value of the ideas of Lavie and Rosenkopf (2006), but does suggest that there might be multiple ways of engaging in ambidexterity. And because the field of ambidexterity at the firm-level is a messy but to a certain extent mature field, I can try to resolve the question how to manage OA by looking at potential synergies or incompatibilities between antecedents at the firm-level, with quantitative methods not applicable in an immature field. The quantitative methods I refer to and why these methods are suitable for this study, will be outlined in the methodology section. Thus, for most firms to realize short-term and long-term success, it is about managing the optimal balance between ET and ER, in which ET and ER compete for the same resources and most likely pull in opposite directions (Simsek et al., 2009). But how exactly managers can carefully manage limited resources to address this trade-off, remains unclear (Simsek et al., 2009; O’Reilly & Tushman, 2013). Currently, the OA literature stream shows evidence for a wide variety of antecedents of both ER and ET at the firm-level, of which several studies hold contradicting results (Junni et al., 2013). Therefore, following the RBV, it is important not to look at these antecedents individually, but to look at them simultaneously in a holistic fashion to explore whether there are potential synergies or compatibilities among these antecedents. Evaluating this empirically can provide interesting insights to address this literature gap, and might help explain why multiple scholars found contradictory findings (O’Reilly & Tushman, 2007, 2013).

3.

HYPOTHESIS DEVELOPMENT

3.1 Learning

interaction of people within an organization, they are often tacit in nature, socially complex and embedded within organizational boundaries. This makes it difficult to imitate learning capabilities (Andreu & Ciborra, 1996).

Learning facilitates ER because it involves the acquisition of knowledge to generate radical ideas (Li et al., 2008; Morgan & Berthon, 2008). On the other hand, firms can learn from its experience to increase its efficiency in current activities, and therefore it influences ET (Li et al., 2008; Wei, Yi & Yuan, 2011). In fact, Wei et al. (2011) argue that learning and ET have a reinforcing effect: when learning helps firms to better understand their current product and process positions, more awareness is provided how to exploit what has currently been done. In turn, firms' competencies in understanding the existing knowledge field grows, which leads to more ET. However, some scholars argue that learning is more related to ER than ET (Morgan & Berthon, 2008); they view the process of learning as the generation of new ideas, which involves outside-the-box thinking. Altogether, I argue that:

H1: A higher level of learning is (a) related positively to a firm’s level of exploitation, and (b) related

positively to its level of exploration, in the presence of the other variables measured in this study.

3.2 Market orientation

Market orientation consists of the generation, dissemination, and responsiveness of market knowledge, which is knowledge focused on customers, competitors and environmental changes (Jaworski & Kohli, 1993). Following the RBV and OL, it is critical to firms to learn to acquire and generate new knowledge and skills, because it will help to keep up or stay ahead of the competition. And in order to learn critical knowledge and skills, it is necessary to solve customers' current and latent needs by closely listening to them (Paladino, 2008). Therefore, market orientation can be considered a capability that contributes to the development of the higher-order organizational learning (Santos-Vijande, Sanzo-Pérez, Álvarez-González & Vázquez-Casielles, 2005). However, some scholars argue, that even though market orientation plays a role in the development of a sustained competitive advantage, it has no direct relationship with superior performance. It requires capitalizing on the market orientation to allow for superior performance (Ketchen, Hult & Slater, 2007). Also, it is how the knowledge generated and disseminated by market orientation is coordinated and integrated via capabilities, then translated into strategic actions, before it can create a sustainable competitive advantage (Murray, Gao & Kotabe, 2011).

there, and therefore often insufficient for breakthrough innovations (Morgan & Berthon, 2008). Proactive market orientation, however, is the attempt to gain in-depth understanding of latent needs, which requires a firm to experiment with unexplored opportunities (Chen et al., 2012). Increased experience in doing so will increase ER (Morgan & Berthon, 2008).

H2: A higher level of market orientation is (a) related positively to a firm’s level of exploitation, and (b)

related positively to its level of exploration, in the presence of the other variables measured in this study.

3.3 Slack resources

Slack resources refer to the difference between the total amount of available resources minus the minimum required resources to produce its organizational output (Cyert & March, 1963). According to the RBV, a firm is able to sustain a competitive advantage when it is able to use its strengths, mitigate its weaknesses, seize opportunities, and avoid threats (Barney, 1991). The ability to do so, is to a certain extent dependent on a firm’s slack resources, because a firm needs some room to manoeuvre (Huang & Li, 2012).

Slack resources are both related to exploratory and exploitative learning. First, they can be deployed to experiment and attempt to seize new business opportunities (Levinthal & March, 1981). For example, in the face of uncertainty, slack resources are associated with the mitigation of control, because they allow an organization to make mistakes, and therefore, facilitate experimentation (Nohria & Gulati, 1996). Second, slack resources can be deployed to upgrade what has currently been done (Meyer, 1982).

H3: A higher level of resources is (a) related positively to a firm’s level of exploitation, and (b)

related positively to its level of exploration, in the presence of the other variables measured in this study.

3.4 Coordination mechanisms

In order to innovate, key knowledge and resources have to be combined into value creating processes that result in either ET or ER (Jansen et al., 2006). Knowledge is often tacit and complex in nature, it is hard to transfer and integrate knowledge into these value creating processes. Therefore, to facilitate knowledge transfer and integration, coordination mechanisms are required (Grant, 1996). Thus, the mechanisms of coordination are not necessarily VRIN resources themselves, but firms use these mechanisms to facilitate innovation, because it eases interaction, integration, and coordination within the organization. This study considers two types of coordination mechanisms: (1) formal, including centralization, formalization, and output control; and (2) informal, including connectedness. Both formal and informal have different effects on ET and ER, as will be outlined next.

uncertainty, because it requires a lower amount of variation, and therefore allows for higher specialization and efficiency of information-processing and decision-making processes. Centralization increases the speed of collective implementation, because constraints are minimal without the necessity of consensus. Furthermore, centralization reduces the costs of problem solving because nearby solutions cost less to develop1 _(Sheremata,

2000). And in turn, is associated with a higher amount of ET (Cardinal, 2001; Jansen et al., 2006).

H4: A higher level of centralization is (a) positively related to a firm’s level of exploitation, and (b) negatively related to its level of exploration, in the presence of the other variables measured in this study.

Formalization. Formalization is defined by Schminke, Ambrose & Cropanzano (2000) as the extent to which the rights and duties of the members of the organization are determined, and written down in rules, procedures, and instructions. Formalization is established to deal with certain circumstances in a known way, that have often been proven to work effectively. Codified practices are easier and faster to implement, which increases the organizational efficiency and control (Zander & Kogut, 1995). Following OL, they are proven to be well-suited for the enhancement of existing resources, and thus, ET (Benner & Tushman, 2003; Jansen et al., 2006).

However, these efficiency gains tend to stifle creativity and reduce flexibility, and ⎯ since creative processes emphasize deviation from existing knowledge and require informal networks to encourage improvisation (Dewett & Williams 2007) ⎯ in turn limit ER (Jansen et al., 2006),

H5: A higher level of formalization is (a) related positively to a firm’s level of exploitation, and (b) related negatively to its level of exploration, in the presence of the other variables measured in this study.

Connectedness. Connectedness is defined as the strength of the linkages within different functional units (Jansen et al., 2006). Knowledge can be difficult to create and transfer within the firm and is often referred to as sticky, since it can be complex and tacit. To transfer sticky knowledge, face-to-face contact is required, which is more likely to occur in firms with higher levels of connectedness. Though, current literature did not reach consensus yet how connectedness influences ET and ER, and results are contradicting (e.g. Atuahene-Gima, 2005; Germain, 1996; Jansen et al., 2006; Koberg, Detienne & Heppard, 2003; Subramaniam & Youndt, 2005). However, I expect that a firm’s connectedness influences both ET and ER, because ET requires extensive sharing of deep knowledge and ER requires extensive sharing of broad knowledge. Following the reasoning of Atuahene-Gima (2005), connectedness builds interfunctional trust, necessary to integrate various interfunctional knowledge for ER (Atuahene-Gima, 2005). Similar for ET, connectedness develops collaboration and trust, which allows for deep understanding to improve existing resources (Atuahene-Gima, 2005; Jansen et al., 2006). Therefore, connectedness can ease the exchange, use and integration of existing and new knowledge and resources (Atuahene-Gima, 2005), and in turn, warrants better decisions about improving existing capabilities

1 _{Only when solving the problem is feasible with searching for solutions nearby old solutions (Sheremata, 2000).}

and creating new ones (Zahra, Ireland, & Hitt, 2000). Consequently, creates an environment favourable for both incremental and radical innovation (Brown & Eisenhardt, 2007). Therefore I hypothesize:

H6: A higher level of connectedness is (a) related positively to a firm’s level of exploitation, and (b)

related positively to its level of exploration, in the presence of the other variables measured in this study.

Output control. Knowledge, and the capacity to create it through organizational learning, is a key resource in terms of value creation and competitive advantage (Grant, 1996). It is argued that individual learning of employees in an organization, is a critical building block for organizational learning and innovation (Argyris & Schön, 1978). Kim (1993) adds that control systems help transforming individual learning into organizational learning. He states that this transformation process is important for innovation, due to potential conflicts that arise from converging different goals associated with the two types of learning. Even with the right employees, goal congruence is difficult to achieve. Therefore, to make sure the employees learn in the direction as stated in the organization's goals, it is important to have output controls in place (Abernethy, Schulz & Bell, 2007). Output control refers to a process that monitors and compensates employees for achieving expected objectives (e.g.: deadlines, performance requirements). Hence, the focus lies on the end result instead of the process towards it. More specific, output control is a construct developed originally by Cardinal (2001), which consists of goal specificity, rewards and recognition, and emphasis on (professional) output.

Control can help managers to make accurate measurements, provide feedback, and helps sharing of knowledge among departments (Simmons,

1991). In general, control involves monitoring deviations from standards and rules, which can be detrimental to creative processes, because creativity requires deviation from the standard (Ford, 1996). Although output control does not require monitoring the process, it does guide the organization towards a beforehand determined objective (Abernethy et al., 2007). Another important aspect of output control is the shift of performance risk from firm to employees, which increases employees’ risk aversion. Consequently, this encourages employees to prefer well-defined behaviours associated with the more predictable ET over trial-and-error learning, associated with the more risky ER (Cardinal, 2001). Therefore I posit the following:

H7: A higher level of output control is (a) related positively to a firm’s level of exploitation, and (b)

related negatively to its level of exploration, in the presence of the other variables measured in this study.

3.5 Synthesis: potential synergies or incompatibilities

The former hypotheses are all measured in a dyadic fashion (Figure 1). However, the essence of this study is an attempt to resolve the mixed findings considering the ambidexterity dilemma. Therefore, it is essential to look at it from a holistic perspective and examine whether these antecedents altogether might yield synergies or incompatibilities. One study revealed that synergies are likely to play a role in OA (Akdoğan, Akdoğan & Cingöz, 2009). They showed evidence that some coordination mechanisms only allow organizations to pursuit ER, ET and their interaction (ambidexterity) when they were combined.2

However, this is a single study and doesn’t encompass most of the antecedents expected to influence OA. Nonetheless, it indicates that potential synergies do exist. Due to the fact that the current ambidexterity literature represents a ‘messy field’ with a wide variety of antecedents with contradictory effects (Junni et al., 2013), a holistic view might be able to grasp the complexity of how the opposing forces of ET and ER can be managed to engage in OA. What synergies will arise and to what extent is still unclear, because prior research has not yet looked at that many variables simultaneously, nor their relative impact on ET and ER. Hence, this part of the research remains in exploratory fashion. Findings can be used as building block for future research to receive further empirical scrutiny and theoretical explanation.

4.

METHODOLOGY

4.1 Data collection

To depict a holistic view, how to manage the balance between ET and ER, it is important to identify important antecedents that either influence ET or ER or both. Many studies have tried to capture these antecedents of OA (Brion, Mothe & Sabatier, 2010; Cao et al., 2009; Jansen et al., 2006; Subramaniam & Youndt, 2005), but the results often were contradicting and fragmented with regard to the type of the antecedents and the nature of the relationship. One of the reasons is that OA has been measured differently by many researchers (Junni et al., 2013). Three meta-analyses have been carried out recently, to resolve these contradictions and capture all relevant antecedents of ET and ER in the literature field (De Boer, 2014; Gubler, 2012; Telman, 2012). A substantial amount of their results correspond with each other. However, due to methodological constraints, they also provide some mixed results3 _{(Appendix A). Nonetheless, this study takes three prior meta-analyses as}

2_{They argue that decentralization of decision-making supports a firm’s exploratory innovations, but without formal and densely connected} structures these new opportunities may not be exploited successfully. Therefore, interactions between these factors that reinforce each other are necessary to achieve ambidexterity (Akdoğan et al., 2009; Jansen, Volberda & Van den Bosch, 2005: 354).

starting point to identify the relevant ambidexterity antecedents, because these studies aimed at identifying relevant organizational antecedents. However, some variables used in these meta-analyses have been left out, due to an insufficient quantity of evidence in the literature. Hence, it was not possible to link these variables to all the independent variables. This simplifies the model to a certain extent, but is not problematic. Because first, a model is a simplification or idealization of the real world, which therefore cannot comprehend every aspect of the real world. Second, it still serves as a building block for a more complex and realistic model. Because these meta-analyses (De Boer, 2014; Gubler 2012; Telman, 2012) only revealed the relationships between the antecedents and ET and ER, and not the relationship these antecedents have among each other, they are insufficient to reveal potential synergies and incompatibilities and the relative importance of these antecedents. Therefore, additional papers had to be collected. This data has been collected in an overarching correlation matrix. Which data is exactly necessary to conduct the analysis and how this exactly has been done will be explained in depth shortly after.

The strategy to search, collect and organize data used in this study followed the basic principles of the book Applied Meta-Analysis for Social Science (Card, 2015) and the guidelines of the Cochrane Handbook (Higgins & Green, 2008). The aim is to be able to draw generalizable conclusions about a research field that has mixed findings and uses heterogeneous samples. Therefore, broad search criteria has been used to find most of the nuances in the OA literature, only, with strict data conditions to allow for the calculations. However, because not all of the relations among the independent variables have been widely researched, I decided to stagnate the search process when I reached at minimum four different effect sizes for every relationship. Although I want to include every relevant study, eventually I reached a point of diminishing returns, where it takes a huge effort to generate only very little extra benefits. At this point, I concluded that it would yield more benefits if I devoted my time and effort into the timely completion of the analysis and development of the optimization method. This means my data collection might not be close enough to exhaustive, therefore I will consider it as a limitation.

Several keywords have been used to acquire the necessary papers, which are all related to the multidimensional constructs of interest. To ensure the reliability of the search process, to make sure studies will reveal a dyadic relationship between the constructs of interest, and to make it more likely that these studies will contain a correlation matrix, every search attempt consisted of a combination of three keywords. Two of these keywords are related to either construct regarding their dyadic relationship, and the third keyword “correlation”4_.

The latter keyword is not sufficient to yield a correlation matrix, however, it is a necessary condition. In general, the keywords chosen are specific to the construct of interest, and as much as possible cover the range of these constructs. In some cases, due to insufficient literature available, multiple keywords covering the subdimensions of these multidimensional constructs have been used to find relevant correlations.5

Some of the organizational antecedents that I have tried to capture, can be measured by scales that have been widely acknowledged and adopted, such as Centralization, Formalization and Market orientation (Jaworski & Kohli, 1993). Though, it has to be mentioned, that most of the studies have their own way of measuring variables, although they define the variable in the same manner. Also, I have tried to capture some constructs of which the field did not yet reach consensus on the exact definition and/or measurement. These studies measure slightly different variables to measure a construct for which they use the same definition (e.g.: Connectedness or

4_{An illustration how it is done in Google Scholar: “CONSTRUCT A” AND “CONSTRUCT B” AND “Correlation.”}

Learning). In order to complete the correlation matrix, I chose to adopt these slightly different variables to measure overarching constructs, such as Connectedness and Learning. Although this method has been adopted more often (Kirca, Jayachandran & Bearden, 2005), this can induce the so-called apples and oranges problem (Card, 2015). This refers to the problem that a researcher tries to capture a certain construct (e.g.: apple) by measuring multiple variables not necessarily the same (e.g.: apples and oranges). To circumvent this problem, the researcher should make sure that the overarching construct being measured covers both the apples and the oranges, and therefore defines its overarching construct as 'apples and oranges' (not 'fruit') (Card, 2015). Therefore, at the measurements section of this methodology, I clearly state which variables I cluster to capture the constructs measured in this study. However, this does not detract from the fact that I deal with effect sizes of potentially heterogeneous constructs, and thus, there might be high variance among them. Even though this is a limitation, this approach has also several advantages. First, by not only using equivalent measurements of constructs, the research can be broadened and adopt a holistic view (Edwards, 2001). If I would have focused solely on similar variables measured in a similar fashion, only a few variables could have been adopted, and these would be too specific to capture the full breadth of some of the (multidimensional) constructs used in this study (Edwards, 2001), and in turn, it would not be impossible to explore potential synergies. Especially considering the contradicting findings in the OA literature, a holistic approach might be necessary to resolve the ET-ER dilemma. Second, studies are seldom identical reproductions of each other, however, including papers that have a distinct methodology, sample and measurements has the benefit of enhancing the generalizability of the conclusions (Rosenthal & DiMatteo, 2001). Lastly, a quantitative way of drawing generalizations by comparing studies, is probably more objective and accurate than trying to solve this dilemma in a narrative fashion (Cooper & Rosenthal, 1980). Therefore, interesting findings might not necessarily represent 'the truth', but do suggest a direction to further research the ET-ER dilemma in a similar, but more extensive fashion.

All the collected effect sizes were combined into one correlation matrix, that represents all the variables adopted in this study. For the completion of the correlation matrix, it is necessary to average the effect sizes of the included papers that examine equivalent relationships. This has been done by first correcting for their sample size, followed up by the Fisher’s Z transformation (Corey, Dunlap & Burke, 1998). This means that the correlation coefficients were firstly being transformed to their associated Fisher's Z, with the following formula:

𝑧 = 0.5 ⋅ ln

!!!_!!!

Then the Fisher's z summary statistic was calculated, and in turn, these statistics were converted back to correlations using:

𝑟 =

!_!!!_!!!!_!!

Even though some argue that Fisher’s Z is biased positively, it is in fact less biased positively, than it is biased negatively (Silver & Dunlap, 1987).

4.2 Sample

This search strategy yielded in total 536 papers. These were all sorted in Excel by author names, journal name, year of publication, title, and the construct(s) of interest and their descriptive statistics. Based on inclusion criteria (Table 1) from 536 studies, 456 were left out due to missing data, which resulted in a final sample size of 80 studies used for the analysis. Due to time constraints, articles of which the necessary data could only be provided upon request, were left out. The use of the search engines Google Scholar and SmartCat include prominent as well as lesser-known publishers, and are not discipline specific. This search strategy resulted in an unbiased sample of studies. Unfortunately, Google Scholar also includes low quality studies. Therefore, to filter the data, only peer-reviewed articles were selected. Also underpowered studies are included, because they were corrected for their sample size. Therefore their influence remains limited.

4.3 Measurements

How each variable is measured in this study is outlined next. An overview of the (multidimensional) constructs is shown in Table 2. An overview of the definitions of these variables is shown in Appendix B; Appendix C presents an overview of all adopted studies.

Dependent variables

include items focused on the improvement of the efficiency and the attempt lowering the costs of current products and services. Studies selected measured exploration by radical innovation or new product or service positions, which includes items related to the invention and experimentation with new products and services and/or new markets (e.g. Bedford, 2015; Jansen et al., 2006; Koberg, Detienne & Heppard, 2003; Troilo, De Luca & Atuahene-Gima, 2014).

Independent variables

There are two broad approaches to measure Market orientation. One is by the more culture-centered conceptualization of Slater and Narver (1994), which divides market orientation in customer orientation, competitor orientation and interfunctional collaboration. The other approach is by Kohli and Jaworski (1990), which divides it in the generation, dissemination and responsiveness of market knowledge. Although the former is more popular, both are equally viable (Hult, Ketchen & Slater, 2005). Narver et al. (2004) later added the distinction between reactive and proactive market orientation. In this study we focus on all these scales (Table 3) to give a broad picture of how market orientation influences ET, ER and their antecedents, because “the three scales appear to be interchangeable” (Deshpande and Farley, 1998, p. 222).

Learning has been measured by adopting Huber's model (1991), which divides organizational learning in the

acquisition, dissemination and shared interpretation of knowledge and its storing of it, in a firm's organizational memory. Although learning is strongly related to market orientation, it goes beyond the focus on the marketplace (Jiménez-Jimenez et al., 2008). In order to fill the correlation matrix, some of the selected studies measure only a few of the dimensions of organizational learning. For instance, Vorhies, Orr and Bush (2011) use the acquisition, dissemination and shared interpretation dimensions to measure learning; Hult and Ferrell (1997) capture the acquisition and dissemination of knowledge dimensions (Hult & Ferrell, 1997). Some studies divide learning in other dimensions, such as explorative and exploitative learning (Huang & Li, 2012) or incremental and radical learning (Salge & Vera, 2012). Also other conceptualizations have been grouped under learning, such as learning orientation (Kropp, Lindsay & Shoham, 2006; Lin et al., 2008), learning culture (Park, 2011), bottom-up learning (Wei & Yuan, 2011) and generative learning (Morgan & Berthon, 2008), because the scales of these conceptualizations also measure the generation and distribution of knowledge.

2005; Li, Chu & Lin, 2010; Nohria & Gulati, 1996). Some of the selected studies divide slack resources in absorbed and unabsorbed slack (Huang & Li, 2012; Liu & Xie, 2014), of which the former measures the available capacity, human resources and time, and the latter measures the availability of retained earnings, financial resources and debt financing with banks. In this study these scales were combine. Although this does not show a nuanced picture, it is required to execute a broad research. Slack resources is also referred to as discretionary slack (Troilo et al., 2014).

In this study Centralization refers to the extent decision making is located at top levels of the firm and to the extent task authority is hierarchical (Hage & Aiken, 1968). All selected studies operationalized centralisation similarly, of which most studies measured it according to the scale of Hage and Aiken (1968) (e.g.: Cardinal, 2001; Hult & Ferrell, 1997; Jansen et al., 2006), some others used the scale of Jaworski and Kohli (1993) (e.g.: Cadogan, Sundqvist & Salminen, 2005; Green Jr, Inman, Brown & Hillman Willis, 2005).

Formalization refers to the extent tasks and rights are formally codified (Hage & Aiken, 1968). All of

the selected studies use a similar definition for formalization, though, measure it with slightly different scales. Again the scale of Hage and Aiken (1968) is most often employed (e.g.: Akdoğan et al., 2009; Cardinal, 2001; Hult & Ferrell, 1997), but also other scales for formalization have been adopted, for example, the scale developed by Deshpande and Zaltman (1982), was used by Jansen et al. (2006).

Connectedness has been defined as the degree of formal and informal face-to-face contact among

employees across departments (Jaworski & Kohli, 1993: p.56). Most of the selected studies measure this via the connectedness scale developed by Jaworski and Kohli (1993) (e.g.: Cadogan et al., 2005; Jansen et al., 2006, 2009; Germain, 1996; Green Jr et al., 2005; Harris, 2000; Hult & Ferrell, 1997). Although a wide variety of variables have been used to measure connectedness (Table 2), they are all clearly related to the same construct. First, some variables have different labels but measure exactly the same. Germain (1996) and Green Jr et al. (2005) use integrative committees and mechanisms, and Hult and Ferrell (1997) use participative and reflective openness, while they all employed the same scale of connectedness by Jaworski and Kohli (1993). Second, as explained earlier, to be able to fill the correlation matrix and engage in a broad exploring research, studies were adopted that use slightly different but related variables (e.g.: interdepartmental connectedness, cross-functional collaboration, socialization, interfunctional teamwork). However, they all share the same purpose mentioned by Jaworski and Kohli (1993): to foster more face-to-face contact among departments. The variable ‘coordination’ (Pelham, 1995) that I used as well might be the most confusing, because in this study coordination refers to not only connectedness, but also centralization, formalization and output control. Still, Pelham (1995) measures with coordination merely the frequency departments work together.

As mentioned earlier, Output control is originally a construct developed by Cardinal (2001), and is in this study measured in a similar fashion. Therefore, only variables similar to output control, or those that are similar to the dimensions of output control, have been adopted (e.g.: Cadogan et al., 2005; Park, 2011). Cardinal (2001) divides output control in goal specificity, performance outcome expectations, transfer effort performance expectations, rewards and recognition and performance feedback.

Control variables

uncertainty and technological uncertainty according to the scales of Jaworski and Kohli (1993) (e.g. Bedford,

2015; Cadogan et al., 2005; De Luca et al., 2007; Jansen et al., 2006; Langerak, Hultink & Robben, 2004; Troilo et al., 2015). The market uncertainty scale includes items that measure the extent to which customer needs and preferences change rapidly and are unpredictable, and the extent to which competitors' actions are changing rapidly and unpredictable. Technological uncertainty scale includes items that measure the extent to which firms experience difficulties to forecast the technological development in the environment (Jaworski & Kohli, 1993). In other studies different definitions of market uncertainty have been used, such as environmental competitiveness, dynamism, turbulence (Atuahene-Gima, 2005; Bedford, 2015; Jansen, Volberda & Van den Bosch, 2005) or competitive intensity (Wei & Yuan, 2011; Wong & Ellis, 2007; Yannopoulos, Auh & Menguc, 2012). The same goes for technological uncertainty. Some studies adopt the definitions of technological turbulence or dynamism (e.g.: Atuahene-Gima, 2005; Atuahene-Gima, Slater & Olson, 2005; Akgün, Keskin & Byrne, 2012; Farrell, 1999; Li et al., 2008). Though, all these studies attempt to measure the same, and most of them use even the same scales (e.g.: Jaworski & Kohli, 1993). Therefore, they can be used interchangeably with market uncertainty or technological uncertainty, respectively.

Both firm age and size were assessed by a ratio scale. Selected studies adopted the measure of firm age as the log of the number of years the firms have been in existence, and firm size as the log of the number of employees (e.g.: Chen et al., 2012; Glisson & Martin, 1980; Kawakami, Maclachlan & Stringfellow, 2012; Wei, Frankwick & Nguyen, 2012).

4.4 Data analysis methods

To assess the validity of the conceptual model and the hypotheses, it is important that the explanatory power of each independent variable will be assessed over and above the predictive power of each variable in the equation. In other words, the predictive power of a variable will be controlled for by all the other variables in the equation (Pallant, 2005). Because I wanted to reveal potential synergies or incompatibilities among OA and its antecedents, and because I dealt with multiple dependent variables, a regular meta-analysis does not suit well (Becker & Wu, 2007). A regular multiple regression analysis does not suit well either, because the examined model includes a link between both dependent variables. This means that the independent variables might not only have a direct effect on ET (ER), but they might also have an indirect on ET (ER), via ER (ET). Therefore I chose to conduct a path analysis because it is the appropriate method to measure both indirect and direct effects of the independent variables on multiple dependent variables. LISREL is a valid tool to conduct a path analysis and assess the validity of the conceptual model (Figure 1). To assess whether the fit of the model is valid, the following fit indices and corresponding values should be realized: RMSEA (0,05<RMSEA<0,08), Satorra-Bentler Chi-Square (χ2 / df ≈ 1,5), NFI (≥0,85), CFI (≥0,85), SRMR (≤0,08), and GFI (≥0,85) (Hu & Satorra-Bentler, 1999). The general multiple regression equation is given by:

𝑦 = 𝛽_!+ 𝛽_!𝑥_!+ 𝛽_!𝑥_!+ … + 𝛽_!𝑥_!+ 𝜀_!

The following equations correspond with the conceptual model6_:

𝑦!"= 𝛽!+ 𝛽!𝐿𝑁𝐺!+ 𝛽!𝑀𝑂!+ 𝛽!𝑅𝐸𝑆!+ 𝛽!𝐶𝐸𝑁!+ 𝛽!𝐹𝑂𝑅!+ 𝛽!𝐶𝑂𝑁!+ 𝛽!𝑂𝐶!+ 𝛽!𝐴𝐺𝐸!+

𝛽!𝑆𝐼𝑍!+ 𝛽!"𝑀𝑈𝑁!+ 𝛽!!𝑇𝑈𝑁!+ 𝛽!"𝐸𝑅 + 𝜀!

𝑦_!"= 𝛽_!+ 𝛽_!𝐿𝑁𝐺_!+ 𝛽_!𝑀𝑂_!+ 𝛽_!𝑅𝐸𝑆_!+ 𝛽_!𝐶𝐸𝑁_!+ 𝛽_!𝐹𝑂𝑅_!+ 𝛽_!𝐶𝑂𝑁_!+ 𝛽_!𝑂𝐶_!+ 𝛽_!𝐴𝐺𝐸_!+ 𝛽!𝑆𝐼𝑍!+ 𝛽!"𝑀𝑈𝑁!+ 𝛽!!𝑇𝑈𝑁!+ 𝜀!

dis required as input, but to realize as accurate results as possible, it is better to include per variable the average mean and standard deviation. Following Cochrane’s method (Higgins & Green, 2008, p.263), to pool the means and standard deviations, I used the standardized mean difference as summary statistic, which is suitable when studies all assess the same outcome but measure it in a variety of ways. Because LISREL is unable to take into account for each square in the correlation matrix a different sample size, I measured with the lowest sample. This means that the results can be biased negatively.

Multivariate conditional distribution

To increase the practical value of this study, I programmed in addition to the MMRA, a multivariate conditional distribution in MATLAB (Appendix E contains the MATLAB program), which reveals what antecedents in the regression equation are the best predictors that influence both dependent variables simultaneously. For this the following formula is required:

𝛴 =

𝛴

_𝛴

!!

𝛴

!" !"

𝛴

!! , with sizes 𝑞 ⋅ 𝑞 𝑞 ⋅ (𝑁 − 𝑞) 𝑁 − 𝑞 ⋅ 𝑞 𝑁 − 𝑞 ⋅ 𝑁 − 𝑞 And,

𝛴 = 𝛴

_!!

− 𝛴

_!"

𝛴

_!!!!

𝛴

_!"

Here,

𝛴

!! refers to the matrix of

[𝑥

!

𝑥

!

]

,

𝛴

!" refers to the matrix of

[𝑥

!

𝑥

!

]

,

𝛴

!! refers to the matrix

[𝑥

!

𝑥

!

]

, and

𝛴

!! refers to the matrix of

[𝑥

!

𝑥

!

]

, of which

𝑥

! consists of the vector of both dependent variables, and

𝑥

! consists of the vector of the independent variables chosen to include in the regression equation. The sizes are determined by

𝑞

, which refers to the vector of dependent variables, by

𝑁

, which refers to the vector of all variables in the equation, and lastly by

𝑁 − 𝑞

, which refers to the residual, independent variables. And

𝛴

represents a

3𝜎

ellipse of the multivariate distribution. The smaller the surface of the ellipse, the better the predictors. First, it reveals the best predictor, and in a sequence, it adds variables that increase the size of the ellipse the least.

Quadratic programming

min

_!!

!

𝑥

!

_{𝐻𝑥 + 𝑥}

!

_𝑓

And is subject to:

𝐴 ⋅ 𝑥 ≤ 𝑏

𝐴

!"

⋅ 𝑥 = 𝑏

!"

𝑙𝑏 ≤ 𝑥 ≤ 𝑢𝑏

In this equation

𝑥

is the unknown, and once solved, represents a vector of all values that represent the smallest necessary adjustments for the independent variables in the regression equation.

𝑥

! _{is the transposed vector of}

_𝑥

(necessary to transform the row vector of

𝑥

into a column vector). Here,

𝐻

,

𝐴

,

𝐴

!" are matrices,

𝑏

and

𝑏

!" are vectors, and

𝑓

,

𝑙𝑏

and

𝑢𝑏

can be both vectors or matrices. Here, optimization deals with capacity constraints. This means that variables should not be able to exceed the minimum value of -1 (

𝑢𝑏

) or the maximum value of 1 (

𝑙𝑏

), because if a variable already reached its maximum value, it is not possible to increase it further. Therefore, the values of the variables entered in the equation should be standardized between values of -1 and 1. The minimum value of -1 corresponds with a 0 on a Likert scale, while 1 corresponds with a 7. Hence, the value of 0 corresponds with an intermediate value of 3.5. Since it might not be realistic to realize values of 0 and 1, organizations can as they please modify the lower and upper bounds (e.g.: 2 to 6, instead of 1 to 7). To minimize

𝑥

for a multivariate conditional distribution, the formulas have to be plugged in into the prior equation. I already mentioned some above, but an additional few are needed:

𝜇 = 𝜇

_!

+ 𝛴

_!"

𝛴

_!! !!

_{(𝑎 − 𝜇}

!

)

Here, the distribution of

𝑥

_! conditional on

𝑥

_!, is

𝑎

, and is required to be multivariate normally distributed, i.e.:

𝑥

!

𝑥

!

= 𝑎 ∼ 𝑁(𝜇, 𝛴)

.

𝜇

represents the desired value of ET and ER (e.g.: the vector [0 0] represents a perfect balance between ET and ER),

𝜇

! refers to the current value of ET and ER (e.g.: [-.5 . 5], which represents 25%-75%),

𝑎

refers to the desired values for the predicting variables to reach the desired value of ET and ER, which is unknown, and

𝜇

! refers to the current value of the predicting variables. Organizations can examine these current values by surveys based on the scales explained in the measurements section.

𝛴

!"

𝛴

!! !! refers to the regression matrix. To test the validity of the equation, a regression analysis via both MATLAB and LISREL has been conducted and the results have been compared with each other. Both programs yielded identical results. So,

𝑎

is the unknown, and to be able to minimize it, the latter formula has to be algebraically transmuted before it can be plugged into quadratic programming:

Here,

𝛥𝜇

_! is

𝜇 − 𝜇

_!, which represents the vector of the required change in ET and ER to realize the desired value of ET and ER (e.g. if

𝜇

is [0 0], and

𝜇

_! is [-.5 .5], then

𝛥𝜇

_! is [.5 -.5]). And 𝛥𝜇! is

𝑎 − 𝜇

!, which represents the vector of the required change in the independent variables to realize

𝛥𝜇

_!, of which only

𝜇

_! is known (e.g.: [.1 .4 .6 -.4 .4 .7 .8 .3 .1]). To find the smallest change in 𝑎 to realize the desired values of ET and ER,

𝛥𝜇

! has to be minimized in absolute sense. By transmuting the former data to quadratic programming, it resulted into the following equation:

min

!!_!

𝑥

!

𝐼

!"

+ 𝑥

!

𝑓

Such that:

[] ⋅ 𝑥 ≤ []

𝛴

_!"

𝛴

_!! !!

_{⋅ 𝑥 = 𝛴}

!"

𝛴

!! !!

𝛥𝜇

!

+ 𝛥𝜇

!

−1 ≤ 𝑥 ≤ 1

Here,

𝐻

is the identity matrix of the total number of variables

𝐼

!".

𝐴

and

𝑏

are empty matrices, and therefore the first constraint can be excluded.

𝐴

!" is the regression matrix,

𝑏

!" is the regression matrix multiplied by

𝛥𝜇

! plus

𝛥𝜇

!, the lower bound is -1 and the upper bound is 1. To run the program, first the program for the multivariate conditional distribution should be run in MATLAB (Appendix E), and subsequently the quadratic program should be run (Appendix F). When both are up and running, fill in the current data for both dependent and independent variables, and the desired value of ET and ER, and run ‘quadprog’.

5.

RESULTS

The correlations among the antecedents and their corresponding means and standard deviations are shown in Table 4. No correlations are exceeding the 0.65 threshold, which suggests our estimations are not likely biased by multicollinearity problems (Tabachnick & Fidell, 1996).

Path analysis model fit. To test hypotheses 1-7, a path analysis was conducted, of which the results are

Test of hypotheses. The positive relationship between learning and exploration (β = .27, p < .001, Model 2)

Hierarchy. The path analysis revealed the

following sequence of importance for the antecedents of ET (Table 6): first, Formalization; second, Output control; third, Market orientation; fourth, Connectedness; fifth, Learning (negative predictor), and sixth, Resources. And the following sequence of importance for antecedents of ER: first, Learning; second, Market orientation; third, Resources; fourth, Centralization (negative predictor); fifth, Connectedness (negative predictor), and sixth, Formalization. The multivariate conditional distribution revealed the following hierarchy of the antecedents that predict both ET and ER simultaneously (Table 6): first, Market orientation; second, Formalization; third, Resources; fourth, Centralization (mainly a negative predictor);