August 20, 2012

August 20, 2012

Ambidexterity: A Meta-Analysis of

Organizational Antecedents

Master Thesis

Business Development

Gerko Telman

s1562347

1

G.A. Telman (s1562347)

Master Thesis MSc BA Business Development

University of Groningen

Ambidexterity: A Meta-Analysis of

Organizational Antecedents

Management summary

The exploitation of existing, mature markets and the exploration of new, emerging markets at the same time is primary factor in the survival and prosperity of a firm. An organization that engages in exploitation and exploration by simultaneously carrying out the paradoxical strategies this calls for is called an ambidextrous organization. The positive influence of ambidexterity performance on firm performance is widely proven and accepted in academic literature. Nonetheless, researchers came to different conclusions about what influences an organization’s ambidexterity performance. The interest of this study is in how an organization’s ambidexterity performance can be managed. The boundaries of this study are set at the firm-level. This research contributes to academic literature by making a first attempt to solve existing contradictions in comprehension of organizational ambidexterity. Meta-analysis was found to be a suitable instrument to produce explanations for the contradictions, because a meta-analysis can solve methodological differences between studies by applying statistical procedures. Using an ambivalent description of ambidexterity, systematic literature search resulted in sixteen studies on organizational ambidexterity providing the data for this study’s meta-analysis. From these sixteen studies, Pearson correlations of eleven organizational antecedents were used as input for meta-analysis. Of the eleven organizational antecedents, only market orientation was found to have a positive, homogeneous significant relation with both tensions of ambidexterity. Output control was found to have a positive, homogeneous significant influence on exploitative innovation. After moderator analysis, non-financial resources also proved to have a positive, homogeneous significant effect on incremental innovation. Learning was found to have a positive, homogeneous significant relation with exploratory innovation. Firm age and structural control had a homogenous but not significant relation with radical innovation. The relations between ambidexterity and centralization, structural connectedness, firm size, formalization and structural complexity were found to be heterogeneous on both exploitation and exploration and remain inconclusive. This study concludes with directions for future research.

Introduction

Most organizations do not survive for long periods of time (Stubbart & Knight, 2006). Agarwal and Gort (1996) showed that firm survival has much to do with the evolution of product cycles. When product cycles progress to maturity, markets stagnate or decline, and players decline along with the market. Levinthal and March (1993) called this the success trap: a focus on exploitation of the market is driving out the exploration of new ones, with potential self-destructing outcomes. On the other hand, engaging in the exploration of new

2

markets may result in an endless cycle of failure, because radical innovations often prove unrewarding or fail altogether – this is called the failure trap (Levinthal & March, 1993). The ability to exploit existing, mature markets and explore new, emerging markets at the same time is primary factor in the survival and prosperity of a firm (March, 1991). An organization that is able to do this is called an ambidextrous organization (Chang & Hughes, 2012; Jansen, van den Bosch & Volberda, 2005; O’Reilly & Tushman, 2011). Many studies have confirmed that organizational ambidexterity positively influences firm performance (Cao, Gedajlovic & Zhang, 2009; Chandrasekaran, Linderman & Schroeder, 2012; Gibson & Birskinshaw, 2004; He & Wong, 2004; Lubatkin, Simsek & Veiga, 2006; McDermott & Prajogo, 2012; Morgan & Berthon, 2008). Organizational ambidexterity is difficult to grasp, because it is achieved by executing paradoxical strategies: the exploitation of markets asks for a focus on efficiency, while exploration asks for experimentation and risk taking (Brion, Mothe & Sabatier, 2010; Benner & Tushman, 2003). In this study, ambidexterity is defined as the ability to manage paradoxical strategies simultaneously. Academic literature provides overwhelming prove of the importance for organizations to be ambidextrous, but empirical results do not provide clarity on what organizational factors influence a firm’s ambidexterity performance, or how these factors should be managed for optimal results.

Aim of the Study

The interest of this study is in how an organization’s ambidexterity performance can be managed. Therefore, the boundaries of this research are set at ambidexterity at the firm-level. The research objective from a theoretical perspective is to recapitulate all previous research on organizational ambidexterity. To the researcher’s knowledge, such a comprehensive, quantitative recapitulation on this subject has not been done before. This research contributes to academic literature by making a first attempt to solve existing contradictions in comprehension of organizational ambidexterity. This article attempts to make the following contributions to organizational ambidexterity literature:

1) This study identifies organizational antecedents of ambidexterity in current academic literature.

2) It tests by means of quantitative evaluation the extent to which organizational antecedents are homogeneous and significant influencers of ambidexterity.

3) It identifies and, if possible, tests moderators for organizational antecedents of ambidexterity that this study found to be heterogeneous after quantitative evaluation.

4) It provides theoretical and managerial implications on this study’s findings.

3

This research tries to solve contradictions in academic literature on the subject of organizational ambidexterity. The contradictions in academic literature may result from methodological problems, differences in study design, different measurements, omitted variables in regression models, and incomparable samples (Song, Podoynitsyna, van der Bij & Halman, 2008). Meta-analysis is a suitable method to resolve these problems, because it operates independently from model composition (Hunter and Schmidt, 1990, 2004; Song et al., 2008). By means of meta-analysis this study integrates previous research on organizational ambidexterity.

This article starts off with a theoretical background. After that, the data collection and methodology are explained. Then the results of the meta-analyses and moderator analyses are presented, followed by conclusions and implications. Lastly, limitations of this study and future research directions are given.

Theoretical background

Over the last decades, much research has been conducted in the field of organizational ambidexterity. The positive relation between ambidexterity performance and firm performance is widely accepted. Nonetheless, researchers came to different conclusions about what influences an organization’s ambidexterity performance. For example, Chang and Hughes (2012), Jansen, van den Bosch and Volberda (2006) and Walker (2008) proved in their studies that high structural complexity, low formalization and low centralization boost exploration, and inverse conditions exploitation, while studies of Cardinal (2001), Dewar and Dutton (1986) and Ettlie, Bridges and O’Keefe (1984) contradicted these results.

Literature research revealed many different antecedents of ambidexterity. Looking closely to definitions and measurements, this study combined the organizational antecedents of ambidexterity into sixteen factors. A detailed description of the factors is given in the following paragraphs, as well as some of the contradictions this study tries to shed light on.

Organizational Antecedents of Ambidexterity

1) Centralization reflects the locus of decision making authority. There is a consensus that high levels of centralization work better for exploitative innovations, while decentralized organizations are more organic and flexible, which positively influences exploration (Boumgarden, Nickerson & Zenger, 2012; Germain, 1996; Grover & Goslar, 1993; Ireland & Webb, 2007; Jansen et al., 2006; Pierce & Delbecq, 1977; Raisch & Birkinshaw, 2008). On the contrary, Ettlie et al. (1984) found a more negative relation between centralization and exploitative innovation, indicating that centralization inhibited exploitation more than exploration. Cardinal (2001) found positive relations of centralization with both exploitative and exploratory innovation.

4

2) Structural connectedness refers to the extent to which employees are networked through the organization. It affects knowledge transfer and learning between employees (Jansen et al., 2006). Where some scholars put structural connectedness as a positive factor for both incremental and radical innovation (e.g. Jansen et al., 2006; Koberg, Detienne & Heppard, 2003; Subramaniam & Youndt, 2005), other academics proved structural connectedness had indeed a positive influence on exploration, but a negative relation with exploitation (e.g. Beckman, 2006; Vanhaverbeke, Gilsing, Beerkens & Duysters, 2009).

3) Firm age has been the subject of debate for a long time (Vanhaverbeke et al., 2009). Some authors address the important role of young entrepreneurial companies in the innovation process, while others assign the same essential role in the innovation process to older, efficient companies (Vanhaverbeke et al., 2009). For example, research of Wei, Yi and Yan (2011) backs up the latter, stating that older companies benefit from more resources and experience, while Jansen et al. (2006) argues that older units may encounter problems in keeping abreast of external developments.

4) Firm size has also been at discussion for its role in the innovation process. Large companies are said to benefit from greater slack and economies of scale, while small companies are advantaged by a greater flexibility (Grover & Goslar, 1993). In an earlier meta-analysis, positive relations between firm size and exploration and exploitation were found, but not with irrefutable evidence (Camisón-Zornoza, Lapiedra- Alcamí, Segarra-Ciprés and Boronat-Navarro, 2004). Firm size is measured in either employees (e.g. Cardinal, 2001; Beckman, 2006; Greve, 2007) or in revenues (e.g. Vanhaverbeke et al., 2009; Wei et al., 2011).

5) Formalization, or the degree to which rules, procedures, instructions and communications are formalized or written down, is in academic literature commonly said to have a positive effect on exploitation, and a negative effect on exploration (Jansen et al., 2006; Menguc & Auh, 2010). Strikingly, Cardinal (2001) found prove for an opposite effect: in her study, formalization was found to have a negative influence on incremental innovation and a positive influence on radical innovation. In accordance with this, results in academic research on the relations between formalization and incremental and radical innovation show mixed results.

6) Learning is defined as the cyclic process of the generation of new distinctions and ideas, the distribution and interpretation of these ideas, and organization’s risk-taking action, and is because of its proactive nature more strongly linked to exploration than to exploitation (Morgan & Berthon, 2008). Mom, van den Bosch and Volberda (2007) agree, stating that learning adds new knowledge to the recipient’s existing knowledge base, thereby enhancing an organization’s exploration performance. On the other hand, Wei et al. (2011) link

5

learning more strongly to exploitation, because organizational specialization makes that employees narrow their search for new distinctions and ideas around their job.

7) Market orientation is an organization’s acquisition, distribution and interpretation of market information, and the organization’s response to it. Morgan and Berthon (2008) argue that market orientation leads to an exploitative innovation strategy because of its responsiveness to the market, failing to take latent customer needs into account. Other scholars disagree with this view, stating that market orientation has a positive effect on both types of innovation (e.g. Li, Lin and Chu, 2008).

8) Output control consists of performance appraisals for quantity of ideas, patents, publications and products, and the rewards received for them. Output control pressures the organization to capitalize on expenditures. Ouchi (1977) argued that output control inadvertently encourages exploitation, because incremental innovation provides organizations with faster returns. Output controls can expected to be detrimental to radical projects (Cardinal, 2001).

9) Resources refers to the level of available resources in three categories: financial resources, human resources and technological resources. All types of resources are needed for both exploration and exploitation, and thus resource constrains force organizations to make tradeoffs between the two (Levinthal & March, 1993). Slack resources provide organizations with greater opportunities to engage in radical innovation, and resources are therefore stronger related to exploration than to exploitation (Greve, 2007). Other studies disproved this by also finding strong positive relations between resources and exploitative innovation (e.g. Ettlie et al., 1984; Quintana-Garcia and Benavides-Velasco, 2007; Wei et al., 2011).

10) Structural complexity is a reflection of the number of functional areas in an organization. It increases the depth of the knowledge base, which in turn increases the development of new ideas (Aiken & Hage, 1971). Studies of Kimberly and Evanisko (1981) and Damanpour (1996) showed significant effects of structural complexity on radical innovations, but not on incremental innovations. Ettlie et al. (1984) found positive effects of structural complexity on both types of innovation.

11) Structural control involves the regulation of activities and behavior by rules and procedures, which, according to Ettlie et al. (1984), supports the development of pre-innovation conditions necessary for exploratory innovation. Cardinal (2001) disagrees by stating that structural control interferes with research activities and may reduce the likelihood that R&D professionals will pursue radical changes that involve higher probabilities of failure.

6

Antecedents outside the scope of this study

In addition to the above mentioned factors, this study found five more organizational antecedents of ambidexterity. For methodological reasons that will be explained later, these five factors are outside the scope of this study. The factors will be discussed briefly, because despite the absence of deeper investigation in this study, they might have influence on an organization’s ambidexterity performance. A recapitulation on the five factors reported here follows in the future research directions.

12) Cultural connectedness refers to the degree in which a shared organizational culture provides employees with a common frame of reference (Beckman, 2006). Shared cultural understandings allow employees to act quickly, causing them to engage in exploitation, while a more diverse cultural understanding implies that employees have more unique ideas and contacts, causing them to engage in exploration (Beckman, 2006).

13) Managerial tenure is associated with performance conformity and strategic persistence if tenures are high, while short tenures are linked with less commitment to the status quo and risky strategies (Koberg et al., 2003). Long managerial tenures are therefore associated with exploitation, and short managerial tenures with exploration.

14) Risk is referred to as the capability of managers to resolve conflicting tensions that occur when making exploration and exploitation decisions (Chandrasekaran et al., 2012). The risk decision capability makes it more likely organizations do not fall into the success trap (exploitation at the expense of exploration) or the failure trap (exploration at the expense of exploitation) (Chandrasekaran et al., 2012).

15) Strategic leadership relates to a contextual alignment capability, which promotes alignment and adaptability across strategic and project levels (Chandrasekaran et al., 2012). Higher levels of alignment promote consistency and clarity of exploitation en exploration goals, and adaptability facilitates the ability to quickly respond to changes in market and customer needs (Chandrasekaran et al., 2012).

16) Structural differentiation allows exploration and exploitation projects to coexist within the same physical setting (Chandrasekaran et al., 2012). However, the presence of dual structures can also lead to isolation and failure of individual units (Simsek, 2009).

7

Data collection

and Methodology

This study performs a quantitative evaluation in the form of meta-analysis in order to solve contradictions that exist in academic literature on organizational ambidexterity. A meta-analysis integrates results of different empirical studies in the same field by applying statistical procedures that solve methodological differences between studies. These statistical procedures are specifically designed for integrating results of empirical studies, allowing it to pool all existing literature on a given topic (Song et al., 2008). A meta-analysis is a suitable instrument to produce explanations for different conclusions between studies.

There are two main types of meta-analysis: the first focusing on a relationship between two variables, the second examining a large number of meta-factors related to one focal construct (Song et al., 2008). The first type of meta-analysis is usually guided by one or two theories, while the second builds on heterogeneous theoretical grounds (Song et al., 2008). This research aims at solving contradictions in literature, building on heterogeneous theories, and thus follows the second type of meta-analysis. Organizational ambidexterity – as the construct of two variables: exploitation and exploration - is used as common denominator, where the main interest of this study is in the range of antecedents reported in academic literature on the subject. Before meta-analysis can be conducted, input for analysis has to be selected, as well as a meta-analytical protocol.

Input for Analysis

The input for this study’s meta-analysis consists of empirical studies on ambidexterity at the firm level. The search for these studies has been conducted in Business Source Premier (BSP), which features full text of more than 2100 journals, dating back to 1965. Ambidexterity has been first mentioned and described in 1976 (Duncan). Consequently, there is no chance of missing articles on the subject older than the contents of the BSP database. The academic literature in the BSP database is systematically scanned for articles on organizational ambidexterity. The concept of ambidexterity is ambivalent, because it comprises the ability to simultaneously manage paradoxical tensions. Therefore, a global description has been used while searching for literature. The global description covers the ability of an organization to simultaneously execute paradoxical strategies regarding innovation. To be able to create an integrated picture of the organizational antecedents of ambidexterity, all studies investigating one or more antecedents in relation with organizational ambidexterity are collected. The BSP database was scanned on peer reviewed journals using the keywords

ambidexterity; exploration combined with exploitation; incremental combined with radical; initiation combined with implementation; transformational combined with modular; transformational combined with modularity; transformational

combined with incremental; discontinuous combined with continuous; breakthrough combined with derivative; disruptive combined with sustaining; and frame breaking combined with incremental.

All collected articles were checked for cross-references to academic literature on organizational ambidexterity. Resulting articles were checked again for cross-references, and so on, until no new studies were found. In

8

total, 62 studies examining the tensions related to ambidexterity were collected. To ensure the studies resulting from the search process were suitable for meta-analysis, three selection criteria were applied: the study had to represent the correct level of analysis, which is ambidexterity at the firm level; the study must significantly reflect the ambidexterity phenomenon; and the study must include a correlation matrix. The correlation matrices contain the data needed as input for meta-analysis. A rather large amount of studies measured ambidexterity imperfectly by integrating the paradoxical tensions underlying ambidexterity into a single variable. The authors of these studies failed to take into account that organizational ambidexterity builds on two basically different tensions. Measuring the concept as a single variable results in unreliable data: the correlations these studies produced are unable to provide insight in how the different variables influenced the paradoxical tensions underlying an organization’s ambidexterity performance. Another restriction on the input for the meta-analysis was that each dataset can only be included once. The number of usable studies was ultimately reduced to sixteen. Appendix 1 shows the selected articles per keyword and per selection criteria. Appendix 2 details the study sample by author, year of publication, sample origin, industry, sample type, sample size and the original independent and dependent variables. Appendix 3 lists the journals in which the studies were originally published.

Meta-analysis: Corrections for Artifacts

The selected protocol for meta-analysis is the one used by Song et al. (2008), based on Hunter and Schmidt (1990, 2004). The protocol makes use of Pearson correlations, which is important because Pearson correlations are independent of other variables in the models that are used as input for this research. In addition, the selected method makes use of random effects models as opposed to fixed effects models. The fixed effects models assume that the same correlation value between meta-factor and dependent variable underlies all studies in the meta-analysis, where random effects models allows for the possibility that population parameters vary from study to study (Song et al., 2008). Given the contradicting relationships between meta-factors and organization ambidexterity across studies, the use of random effects models is appropriate.

The second step is to make corrections for every independent variable or meta-factor related to a dependent variable based on three artifacts: dichotomization, sampling error and error of measurement. Dichotomization reduces the real correlation by at least 0.8 (Hunter and Schmidt, 1990, 2004; Song et al., 2008). The correction for correlations of individual dichotomized variables is as follows:

9

where ad is the correction for dichotomization, which is 0.8 for dichotomized variables and 1 if the variable is

not dichotomized; rooi is the observed correlation of primary study i; and roi is the correlation of primary study

i corrected for dichotomization.

Second, correlations are corrected for sampling error by weighing them by sample size (Hunter and Schmidt, 1990, 2004; Song et al., 2008). The formula for correcting for sampling error is:

r

  ∑ N_{∑ N} r

  

where Ni is the sample size of of primary study i; roi the correlation corrected for dichotomization as obtained

in the previous step; and weighted ro is the correlation of a given meta-factor corrected for dichotomization and sample size.

Third, measurement errors were resolved by making use of the Cronbach’s alphas reported in the primary studies. If Cronbach’s alphas were not mentioned, alphas of the same meta-factors in the other studies were averaged and used as input for the formula. The average correlation corrected for dichotomization and sample size was divided by the product of the average square root of the reliabilities of the meta-factor and the average square root of the reliabilities of the dependent variable. The formula for real population correlation is:

ρ  r_{A }  r R

  R 

where weighted ro is the correlation corrected for dichotomization and sample size; weighted A the

compound reliability correction factor; √Rxx the square root of the reliability of the meta-factor; and √Ryy is

the square root of reliability of the dependent variable.

Meta-analysis: Variance-based Tests

The next step in the meta-analysis protocol is to determine whether a homogeneous relationship exists between a meta-factor and the dependent variables. A chi-square test, often used in other meta-analysis, will not provide a satisfying answer, because it is not corrected for artifacts (Hunter and Schmidt, 1990, 2004; Song et al, 2008). Instead, a variance-based test is suggested. The total variance in the correlation coefficients has three sources: variance due to artifacts (dichotomization and measurement errors), variance due to sampling error, and real variance due to heterogeneity of the meta-factor (Song et al., 2008). If the real variance is no more than 25 percent of the total variance, the meta-factor is assumed to be homogeneous

10

(Hunter and Schmid, 1990, 2004; Song et al., 2008). First, the total variance is calculated. It can be obtained by using the following formula:

Var   ∑ Nr 

  r  !"

∑ N 

where Ni is the sample size of primary study i; rooi the observed correlation of primary study i; and weighted roo

is the average of observed correlations.

Second, the variance due to artifacts is calculated. The variance of reliabilities of the meta-factor is divided by the average square root of reliabilities of the meta-factor, and added to the variance of reliabilities of the dependent variable divided by the average square root of reliabilities of the dependent variable. This outcome is multiplied with the squared averaged sample correlation corrected for artifacts. The formula is as follows:

Var # ρ! A! V  r! V  r! VarR

√R

 % VarR_{√R }

where weighted ro is the sample correlation corrected for artifacts; Rxx is the reliability of the meta-factor; and

Ryy is the reliability of the dependent variable.

Third, the variance due to sampling error is calculated. The weighted average of observed correlations is squared and subtracted of one, squared again and divided by the average sample size of a meta-factor minus one. It is calculated using the following formula:

Var&.(.1  r ! ! N  1 % * +,-ad/1 !  101  r_N ! ! 1 1 2  1  r! ! N  1 % * ,0.5625 1  r ! ! N1 1 0 D 1 D 1

where weighted roo is the average of observed correlations; and weighted N is the average sample size of the

meta-factor. Only the first part of this formula is used, because no dichotomization was observed in this study’s data.

With the output of these formulas the real variance can be calculated, simply by subtracting the variance due to artifacts and the variance due to sampling error from the total variance. Explained variance should be at least 75 percent to yield a homogeneous meta-factor. If real variance is less than 25 percent of the total variance of correlations from the primary studies, it is likely due to unknown and uncorrected artifacts and can therefore be neglected (Song et al., 2008). Two significance test remain for homogeneous meta-factors: a

11

p-value and a confidence interval. The p-value indicates the degree of significance of the correlation. If its value is 0.05 or below, the relation is significant. The p-value is obtained by the following formula:

8  ρ

9 1  ρ_{∑ N} !

 

  2

where p is the real population correlation; and Ni the sample size of primary study i.

The p-value alone is not enough to build conclusions on. A confidence interval is calculated as a second test. When the confidence interval includes zero, the nature of the relationship between meta-factor and dependent variable cannot be determined.

For heterogeneous meta-factors, moderator analysis is applied. This is done by dividing data into subgroups, according to different performance measures, differences between samples, or different meta-factor measures. For each subgroup, a separate analysis was conducted, in order to find homogeneous meta-factors within the subgroups.

The last step in the meta-analysis protocol is to perform a file drawer test, in order to assess publication bias. There is a tendency to only publish significant results, while insignificant studies disappear in file drawers (Song et al., 2008). The file drawer test produces a number, which indicates the amount of null-result studies that when added makes the meta-factor’s total significance exceed the critical level of 0.05. The file drawer is represented by Xs in this study. The higher its value, the more reliable the results, while a Xs of zero indicates that a meta-factor is insignificant. The Xs is calculated using the following formula:

:;  _{2.706  <  ><} < _{! 2.706"}

where K is the number of studies that build a meta-factor; and Z is the Z-value of a given meta-factor.

Analysis and Results

Data collection revealed sixteen organizational antecedents of ambidexterity. Appendix 4 shows the sixteen factors along with the variables reported in the original studies. This study only includes antecedents that are measured at least twice in previous research studies, as to limit the sensitivity of the results. This resulted in eleven organizational antecedents of ambidexterity as input for meta-analysis. Definitions of the eleven antecedents are outlined in table 1. In several cases, one study delivered more than one variable as input for a single meta-factor. For example, Germain (1996) measured interdepartmental connectedness and integration

12

as separate independent variables, but they are both parameters of structural connectedness. When several variables related to one meta-factor were researched within a single study, a decision was made to either average both the correlation and Cronbach alphas, or to include only the independent variable that was most representative of the meta-factor. This decision was based on the definitions and measurements of the variables.

Table 1: Definitions of the eleven Meta-factors

Meta-factor Definition Selected References

Centralization Vertical locus of decision-making authority Germain, 1996 Connectedness (structural) Extent to which employees are networked to

various levels of the hierarchy in the organization

Jansen et al., 2006

Firm Age Firm age in years Vanhaverbeke et al., 2009

Firm Size Firm size in employees Koberg et al., 2003

Formalization Degree to which rules, procedures, instructions and communications are formalized or written down

Jansen et al., 2006

Learning The cyclic process of the generation of new distinctions and ideas, the distribution and interpretation of these ideas, and the organization’s risk-taking action

Morgan and Berthon, 2008

Market Orientation Market information acquisition, distribution, interpretation and response by the organization

Morgan and Berthon, 2008 Output Control Reflection of performance appraisals for quantity

of new ideas, new patents, publications and new products, and the reflection of rewards received for the same parameters

Cardinal, 2001

Resources Level of available resources Li et al., 2008

Structural Complexity Number of functional areas in an organization Jansen et al., 2006 Structural Control Degree to which activities and behaviors are

regulated by rules and procedures

Cardinal, 2001

Organizational ambidexterity is an ambivalent concept of paradoxical tensions. These tensions are most commonly denoted as exploitation and exploration, and henceforth labeled as such. Each tension was subjected to meta-analysis, producing two sets of outcomes. Results of the meta-analysis are presented for exploitation in table 2 and for exploration in table 3. The tables present in following order: the meta-factors; total N or the composite sample size; K or the number of correlations that build a meta-factor; p, which is an estimate of the real population correlation; the 95 percent confidence interval of the real population correlation; the explained variance; the presence of moderators for a given meta-factor; and Xs, the critical number of null-result studies.

13

Table 2: Results of the Meta-analysis for Exploitation

EXPLOITATION Total N K p (1) 95% Confidence Interval Explained Variance (%) (2) Moderators Xs Meta-factor Centralization 579 4 -0.10 (-0.31,0.11) 38 Yes 0 Connectedness (structural) 1008 6 0.17*** (-0.18,0.51) 16 Yes 140

Firm Age 945 5 0.08** (-0.09,0.25) 41 Yes 56

Firm Size 8998 14 -0.17 (-1,0.68) 1 Yes (3) 0

Formalization 825 5 0.32*** (0.16,0.48) 44 Yes 96 Learning 373 2 0.40*** (0.12,0.68) 20 Yes 14 Market Orientation 387 2 0.35*** (0.30,0.39) 91 14 Output Control 113 2 0.19** 100 5 Resources 5875 8 -0.23 (-0.81,0.35) 2 Yes (3) 0 Structural Complexity 3389 3 -0.34 (-0.60,-0.08) 7 Yes 0

Structural Control 113 2 0.14* (-0.15,0.43) 44 Yes 1

(1) For all p-values, one tailed test statistic; direction depends on sign of p; (2) Explained variance below 75% means that the meta-factor has moderators; (3) See table 5 and 6 for moderator analysis

* p < .05 ** p < .01 *** p < .001

Table 3: Results of the Meta-analysis for Exploration

EXPLORATION Total N K p (1) 95% Confidence Interval Explained Variance (%)(2) Moderators Xs Meta-factor Centralization 579 4 -0.47 (-1,0.21) 5 Yes 0 Connectedness (structural) 1008 6 0.16** (-0.07,0.38) 31 Yes 65 Firm Age 945 5 -0.05 100 0

Firm Size 8998 14 -0.28 (-1.0,0.76) 1 Yes (3) 0

Formalization 825 5 0.12* (-0.19,0.43) 19 Yes 19

Learning 373 2 0.40*** 100 14

Market Orientation 387 2 0.35*** 100 14

Output Control 113 2 0.43*** (0.21,0.65) 53 Yes 14

Resources 5875 8 0.41*** (0.34,0.48) 58 Yes (3) 252

Structural Complexity

3389 3 -0.26 (-.50,-0.03) 6 Yes 0

Structural Control 113 2 0.12 100 0

(1) For all p-values, one tailed test statistic; direction depends on sign of p; (2) Explained variance below 75% means that the meta-factor has moderators; (3) See table 5 and 6 for moderator analysis

* p < .05 ** p < .01 *** p < .001

14

The results in table 2 reveal two homogeneous significant organizational antecedents of exploitation, and table 3 reveals two homogeneous significant organizational antecedents of exploration. These are

for exploitation: 1) Market orientation (p = 0.35, p < 0.001) 2) Output control (p = 0.19, p < 0.01)

for exploration: 1) Learning (p = 0.40, p < 0.001)

2) Market orientation (p = 0.35, p < 0.001)

Results in table 3 indicate that firm age and structural control are homogeneously related to exploration, but this relation is not significant. Except for market orientation, all variables proved to have heterogeneous relations with exploration, exploitation, or both. These variables will be treated with next.

Moderators

Table 2 showed that two of eleven meta-factors have homogeneous significant correlations with exploitation, and table 3 showed that four of eleven meta-factors have homogeneous correlations with exploration, of which two were significant. Only market orientation was found to have homogeneous relations with both exploration and exploitation, the ten remaining meta-factors have heterogeneous relations to either exploitation, exploration or both. This implies that their relation to organizational ambidexterity depends on situations. The ten meta-factors with at least one heterogeneous relation with ambidexterity are: centralization, structural connectedness, firm age, firm size, formalization, learning, output control, resources, structural complexity and structural control. All these variables are examined for the presence of subgroups. If a subgroup can be identified, the meta-factor in question will be subjected to moderator analysis, provided that enough data is available. At least two studies are needed as input for each subgroup. Subgroups can be found in differences between performance measures, sample differences or differences in meta-factor measures. For three of ten heterogeneous meta-factors subgroups were identified. First, firm size was measured in two distinct ways, with possibly different outcomes. Second, Wei et al. (2011) argued that specialization causes employees to narrow their learning around their job. Samples in which high specialization was found may therefore report different relations with ambidexterity than samples in which specialization was considerably lower. Therefore, specialization was selected as moderator for learning. Third, the meta-factor resources was measured in distinct ways. Some studies focused on financial resources, others on human resources and one study reported technological resources. The three meta-factors with the type of moderator and the accompanying subgroups are shown in table 4.

15

Table 4: Moderators and Subgroups

Meta-factor Type of moderator Subgroups

Firm Size Meta-factor measures In employees

In revenues

Learning Sample differences High specialization

Low specialization

Resources Meta-factor measures Financial

Non-financial

This study’s dataset provided input for moderator analysis of only two meta-factors: firm size and resources. More research is needed to provide enough input for the moderator analysis of the meta-factor learning.

Table 5 presents the results of moderator analysis for exploitation and table 6 presents the results of moderator analysis for exploration. The real population correlation is noted as p, the aggregate sample size as

N, the number of correlations that build a meta-factor as K, followed by the 95 percent confidence interval,

the explained variance, and Xs, the critical number of null-result studies. Following the rules for meta-analysis, explained variance should be at least 75 percent to yield a homogeneous factor. The two significance tests are also applied to the moderator analysis: a p-value, which should be 0.05 or below, and a confidence interval, which may not include zero. In the moderator analysis, confidence intervals of distinct subgroups are also not allowed to overlap (Song et al., 2008). If confidence intervals of more than one moderator of a meta-factor do overlap, the subgroups may not have a distinct effect on the dependent variable after all.

Table 5: Moderator Analysis for Exploitation

EXPLOITATION Moderator p (1) Total

N K 95% Confidence Interval Explained Variance (%) (2) Xs Meta-factor

Firm Size in employees -0.18 8559 11 (-1,0.69) 1 0

in revenues 0.01 439 3 (-0.44,0.47) 11 0

Resources financial -0.43 4501 3 (-0.73,-0.13) 2 0

non-financial 0.33*** 1374 5 100 96

(1) For all p-values, one tailed test statistic; direction depends on sign of p; (2) Explained variance below 75% means that the meta-factor has moderators; * p < .05

** p < .01 *** p < .001

16

Table 6: Moderator Analysis for Exploration

EXPLORATION Moderator p (1) Total

N K 95% Confidence Interval Explained Variance (%)(2) Xs Meta-factor

Firm Size in employees -0.30 8559 11 (-1,0.76) 0 0

in revenues 0.06 439 3 (-0.12,0.25) 44 0

Resources financial 0.39*** 4501 3 (0.25,0.54) 46 96

non-financial 0.43*** 1374 5 (0.38,0.49) 42 33

(1) For all p-values, one tailed test statistic; direction depends on sign of p; (2) Explained variance below 75% means that the meta-factor has moderators; * p < .05

** p < .01 *** p < .001

Results of the moderator analysis in table 5 and 6 show that only resources had a distinct moderator subgroup: non-financial resources. Non-financial resources was found to be a homogeneous significant moderator of the relation between resources and exploitation. For exploration the relation remained heterogeneous. Financial resources had heterogeneous relations with both exploitation and exploration and was therefore not a moderator of resources. Firm size was not found to have distinct moderator subgroups of any kind, both firm size in employees and in revenues had heterogeneous relations with both exploitation and exploration.

Looking back at table 2 and 3, ten meta-factors proved inconclusive for at least one of both tensions of ambidexterity. For seven of these meta-factors no subgroups could be identified in this study, which made them unsuitable for moderator analysis. These seven meta-factors are: centralization, structural connectedness, firm age, formalization, output control, structural complexity and structural control. Moderator analysis provided proof for a homogeneous relation between non-financial resources and exploitation, but for financial resources the relation remained inconclusive, as well as for both subgroups of resources with exploration. For one of the inconclusive meta-factors a subgroup could be identified, but with less than two studies providing data for moderation. This was the case with learning. Further research is needed to validate or disprove this potential moderator.

Conclusions and Implications

The purpose of this study was to uncover organizational antecedents of ambidexterity by integrating previous research towards the subject. The results of this study are summarized in figure 1. Left and right are the organizational antecedents investigated in this study, which are connected to ambidexterity in the middle. Each variable is depicted twice. The variables on the left side of the model connect to exploitation and the variables on the right side connect to exploration, corresponding with this study’s meta-analyses. Variables

17

with the same relations to exploration or exploitation are grouped. The straight lines present homogeneous relations, the dotted lines present heterogeneous relations. Furthermore, significant positive relations and non-significant relations are indicated. This study did not produce significant negative relations. Of the moderators, only the homogeneous significant are depicted in the figure.

Figure 1: Overview of Relations Meta-factors with Exploitation and Exploration, and Moderators

Major Research Results

Results of this study’s meta-analysis show that there is good reason for the absence of a paradigm on the management of organizational ambidexterity: previous research provided enough data to conduct meta-analysis for only eleven of sixteen organizational antecedents of ambidexterity, and of these eleven antecedents only market orientation was found to have positive, homogeneous significant correlations with both tensions of ambidexterity. An organization’s market orientation is thus an important influencer of organizational ambidexterity performance. For seven antecedents, no proof of a relation with ambidexterity was uncovered at all. The other three antecedents showed to be related with only one tension of the ambidexterity concept: output control and non-financial resources were found to have positive, homogeneous significant relations with exploitation, and learning was found to have a positive, homogeneous significant relation with exploration. The relations of output control, non-financial resources and learning to only one tension of organizational ambidexterity provides some insight to how ambidexterity is constituted in organizations, but the results are not substantial enough to prove these factors are indisputable organizational

Centralization Connectedness (structural) Firm Age Firm Size Formalization Learning Structural Complexity Structural Control Firm Age Structural Control Centralization Connectedness (structural) Firm Size Formalization Output Control Resources Structural Complexity Learning Market Orientation Market Orientation Output Control Resources E xp lo ra tio n E xp lo it at io n Ambidexterity

Non-financial Types of relations: heterogeneous homogeneous n.s. : non-significant + : significant positive - : significant negative n.s. + + +

18

antecedents of the whole concept of ambidexterity. More research is needed for all 10 heterogeneous meta-factors in table 2.

Theoretical and Managerial Implications

An important implication from meta-analysis is that insignificant results may be due to heterogeneous factors (Song et al., 2008). Contradictions among studies on organizational ambidexterity that still exist after meta-analysis may be explained in future studies, investigating differences in correlation coefficients of a independent and a dependent variable between different subsamples. The theoretical and managerial implications are given in the next paragraphs, the first focusing on ambidexterity as the combination of two tensions, then two paragraphs focusing on exploitation and exploration individually, and the last paragraph focusing on organizational factors of which this study’s results indicated they had no relation with ambidexterity.

Ambidexterity

In this meta-analysis, one independent variable was found to influence both tensions of ambidexterity: market orientation. An important conclusion of this study is that an organization’s market orientation is beneficial for its ambidextrous performance, and subsequently for an organization’s overall performance. The meta-analysis in this study revealed a correlation of 0.35 between market orientation and both exploitation and exploration. This implies that incremental and radical innovation performance benefit equally from an organization’s market information acquisition, distribution, interpretation and response to this information.

Exploitative innovation

For some of the other meta-factors in this study, homogeneous significant relations with only one part of ambidexterity were found. These results can be useful when an organization is seeking to enhance either its exploitation or its exploration performance. First, output control was found to have a positive, homogeneous significant effect on exploitative innovation. This implies that performance appraisals, rewards and recognition for quantities of new ideas, patents, products and publications have value in enhancing incremental innovation, but the same might not be true for radical innovation. This finding is in line with the study of Ouchi (1977): pressure to increase output inadvertently encourages incremental innovation.

After moderation, non-financial resources were found to have a positive, homogeneous significant effect on exploitative innovation. This shows that higher levels of human and technological resources positively influence an organization’s incremental innovation performance, while such a relation between financial resources and exploitative innovation was not proved in this study. Human and technological resources often become institutionalized and transferred throughout an organization (Subramaniam & Youndt, 2005), making

19

an organization more efficient. The increased efficiency of an organization might explain the positive relation between human and technological resources and incremental innovation performance, because exploitation asks for a focus on efficiency (Brion et al., 2010, Benner & Tushman, 2003).

Exploratory innovation

Other meta-factors proved to be homogeneously related to only the exploratory side of ambidexterity. Learning was found to have a positive, homogeneous significant effect on radical innovation. Thus, radical innovation benefits from a process of generation of new distinctions and ideas, distribution and interpretation of these distinctions and ideas, and an organization’s risk taking action. This is not a compelling conclusion, because the concept of learning is interrelated with market orientation (Morgan & Berthon, 2008). However, this implies that learning also has a positively influence on exploitative innovation. This study did not found prove for such a relation. Therefore, more research is needed to prove that learning is related with both ends of organizational ambidexterity.

Structural control was found to have an insignificant but homogeneous relation with exploration. Meta-analysis showed a correlation of 0.12, which might be explained by the presence of a curvilinear relationship: too much structural control inhibits radical innovation, while the absence of it has the same effect. Exploratory innovation might be best served by a moderate level of structural control. The other meta-factors all had heterogeneous relations with exploratory innovation, even after moderation. This means more research is needed toward the exploration part of ambidexterity.

Unrelated factors

Firm age proved to have a heterogeneous relation with exploitation, and a homogeneous relation with exploratory innovation, but both relations were insignificant. Both correlations were very close to zero (0.08 for exploitation; -0.05 for exploration). A possible explanation is that for older companies the positive effects resulting from more resources and experience are cancelled out by negative effects of incumbency, implying that firm age has no influence on ambidexterity at all.

Limitations

Despite the corrections for different types of errors in previous research, this study’s meta-analysis also has its weak points. First, the Pearson correlations used are primarily intended for measurement of the strength of a linear relationship between two variables (Song et al., 2008). In the case of curvilinear relationships, Pearson correlations might, incorrectly, report no relation at all. A second limitation was the inclusion of a meta-analysis by Damanpour (1996) in this study’s meta-meta-analysis. The calculations of this meta-meta-analysis were not available and have not been recalculated for this study, nor has the study by Damanpour (1996) been checked

20

by the author of this study for applying correct definitions and measurements on its input. A third limitation was the sample size of the meta-analysis self: it included only 16 studies. Primary reason for the small sample size, while much more research on organizational ambidexterity exists, is that many scholars measured ambidexterity as a single factor, not taking into account that ambidexterity builds on two basically different tensions. For this reason, many studies on organizational ambidexterity were excluded from this study’s meta-analysis. Although the sample size of this study is limited, it is sufficient for a preliminary meta-analysis (Gerwin & Barrowman, 2002; Song et al., 2008).

Future Research Directions

Rather than making future research to organizational ambidexterity superfluous, this meta-analysis stimulates it and provides directions by exposing flaws in current academic literature on the subject. Based on this study’s results, implications and limitations, some directions for future research are given.

Measurement of Ambidexterity

Ambidexterity was defined in this study as the ability to manage paradoxical strategies simultaneously. The concept stretches between two tensions. A high exploration score can cancel out a low exploitation score, et vice versa, thereby biasing a study’s correlations if the concept of ambidexterity is measured as a single variable. Studies investigating ambidexterity as not ambivalent may find significant correlations, but de facto these studies fail to provide practical insights to management of organizational antecedents of ambidexterity. Therefore, one of the most important directions for future research is that ambidexterity should not be investigated as a single variable, but as two variables: one relating to incremental innovations and one relating to radical innovation. Valid, reliable measurement scales can be found in, for example, Chang, Hughes and Hotho (2011), Jansen et al. (2006) and Morgan and Berthon (2008).

Extending the body of knowledge

For some of the organizational antecedents in this study no proof was found for a homogeneous, significant or homogeneous significant relation with ambidexterity. For five of these antecedents, no moderators could be identified. To come to agreement on the influence of these organizational factors on ambidexterity, more research on these factors is needed.

First off, the correlations between the locus of decision making authority – centralization - and incremental and radical innovation performance were found to be homogeneous nor significant. Considering the limited sample size for this meta-factor of four studies, more research is needed before a consensus can be created. Second, only five studies measured formalization in relation with both tensions of ambidexterity. Integrating the results of these studies did not provide substantial proof for the consensus that formalization positively

21

influences exploitation and negatively influences exploration. Both correlations were found to be positive and significant, but not homogeneous. More research is needed to provide substantial proof for the relation between formalization and ambidexterity. Third, this study found a homogeneous significant relation between

output control and incremental innovation, but the relation with radical innovation was heterogeneous - albeit

significant and with reasonably high explained variance at 53 percent. Considering the low composite sample size of 113, this gives reason to believe that future studies might be able to find less challengeable proof of the relation between output control and exploration. Fourth, for structural complexity, studies of Damanpour (1996) and Ettlie et al. (1984) came to different conclusions. Only one other study (Jansen et al., 2006) investigated the relationship between structural complexity and ambidexterity, reporting weak correlations. Logically, with only these three studies as input, this meta-analysis did neither find significant or homogeneous relations. More research is needed to explain the conflicting results in the studies of Damanpour (1996) and Ettlie et al. (1984). Lastly, the relations between structural control and ambidexterity were either heterogeneous or insignificant in this study. The inconclusive results may stem from the composite sample size of 113, which is small for a meta-analysis. Additionally, because of the low correlation this study produced (0.14 for exploitation; 0.12 for exploration), there is reason to suspect a curvilinear relationship between structural control and exploration. Regulation of activities and behavior might indeed support pre-innovation conditions, as supported by Ettlie et al. (1984), but overregulation may undo these advantages. Regression studies on the subject may offer a solution.

Antecedents outside the scope of this study

This study’s data collection provided input for meta-analysis of eleven of sixteen identified antecedents. The remaining five antecedents were only investigated once in previous research for their relation with exploration and exploitation. More research on these organizational antecedents is needed in order to produce a sophisticated view on how these antecedents are ideally managed to enhance an organization’s ambidexterity performance. The five factors are: cultural connectedness, managerial tenure, risk, strategic leadership and structural

differentiation. Outlines of the definitions of these factors are given in this study’s theoretical background.

Moderating Variables

Other variables also proved to have heterogeneous, insignificant or heterogeneous insignificant relations with organizational ambidexterity, but this study found reason to believe the relations between these antecedents and ambidexterity are moderated by other factors. Future studies might yield better results when moderating for these factors.

Firstly, despite significant, positive correlations, connectedness was not homogeneously related to exploitation and exploration. Without a good explanation for the variation in correlations between studies, no consensus

22

can be reached on this subject. It is encouraged to continue research on the relation between connectedness and ambidexterity. Taking into mind that the relations between this meta-factor and both incremental and radical innovation were heterogeneous, structural connectedness may be influenced by moderating variables. A possible moderator might be the extent to which different functional areas are located in the same physical place, because connectedness might benefit from a more centralized physical organization. Second, a positive relation between learning and exploration was proved in this study, but the relation between learning and exploitation remains unclear. The small percentage of explained variance indicated that this might be due to moderators. As proposed by Wei et al. (2011), this study adopts the view that the link between learning and incremental innovation might be moderated by specialization. The studies that provided input for the meta-factor learning did not moderate for specialization, leaving a research opportunity for other scholars. Third, this study found strong proof for a positive relation between market orientation and organizational ambidexterity, which is a valuable conclusion. However, understanding of this relation can be deepened even further. This study’s meta-analysis measured market orientation as the combination of responsive market orientation and pro-active market orientation. This was done because only one of the studies used as input for the meta-analysis measured market orientation as two different concepts. Narver, Slater and MacLachlan (2004) defined responsive market orientation as a business’s attempt to understand and satisfy customers’ expressed needs, and a proactive market orientation as the attempt to understand and satisfy customers’ latent needs. Following these definitions of market orientation, it might be true that incremental innovation benefits more from a responsive market orientation, and radical innovation from a proactive market orientation. This was already proven by Li et al. (2008). This is a useful insight in managing the paradoxical tensions of ambidexterity, and therefore it is proposed that future researchers make a distinction between reactive market orientation and proactive market orientation when investigating its relation with the two tensions of ambidexterity. A high-quality measurement scale for responsive and proactive market orientation can be found in the study of Li et al. (2008). Lastly, strong proof was only found for the relation between non-financial resources and incremental innovation. For non-financial resources and incremental innovation, and for both types of resources with exploratory innovation, the nature of the relation remains unsolved. The significant correlations and reasonably high levels of explained variance indicate that deeper investigation of this moderator might yield convincing proof for the relations of both types of resources with exploitation and exploration.

All in all, this study provided proof for a positive relation between market orientation and ambidexterity, and showed there is much research to be done on other organizational antecedents of ambidexterity before a consensus can be created. This study’s preliminary meta-analysis stimulates further research on the subject of

23

ambidexterity and provides future scholars with research opportunities by giving directions for insufficiently investigated organizational antecedents of ambidexterity, and some moderators.

Acknowledgements

The author of this study thanks Dr. J.D. van der Bij and Dr. C. Reezigt, both associate professors of the department of Innovation Management & Strategy at the faculty of Economics and Business at the university of Groningen, for their valuable comments and suggestions for improvement.

24

References

• Agarwal, R. & Gort, M. 1996. The Evolution of Markets and Entry, Exit, and Survival of Firms.

Review of Economics & Statistics, 78(3): 489-498.

• Aiken, M. & Hage, J. 1971. The Organic Organization & Innovation. Sociology, 5(1): 63-82.

• Beckman, C.M. 2006. The Influence of Founding Team Company Affiliations on Firm Behavior.

Academy of Management Journal, 49(4): 741-758.

• Benner, M.J. & Tushman, M.L. 2003. Exploitation, Exploration, and Process Management: The Productivity Dilemma Revisited. Academy of Management Review, 28(2): 238-256.

• Boumgarden, P., Nickerson, J. & Zenger, T.R. 2012. Sailing into the Wind: Exploring the Relationships among Ambidexterity, Vacillation, and Organizational Performance. Strategic

Management Journal, 33(6): 587-610.

• Brion, S., Mothe, C. & Sabatier, M. 2010. The Impact of Organisational Context and Competences on Innovation Ambidexterity. International Journal of Innovation Management, 14(2): 151-178.

• Business Source Premier. Beschikbaar: http://www.ebscohost.com/academic/business-source-premier.

• Cabello-Medina, C., Carmona-Lavado, A. & Cuevas-Rodríguez, G. 2011. Organisation of R&D Departments as Determinant of Innovation: Testing Alternative Explanations. Technology Analysis &

Strategic Management, 23(4): 383-400.

• Camisón- Zornoza, C., Lapiedra- Alcamí, R., Segarra-Ciprés, M. & Boronat-Navarro, M. 2004. A Meta-Analysis of Innovation and Organizational Size. Organization Studies, 25(3): 331-361.

• Cao, Q., Gedajlovic, E. & Zhang, H. 2009. Unpacking Organizational Ambidexterity: Dimensions, Contingencies, and Synergistic Effects. Organization Science, 20(4): 781-796.

• Cardinal, L.B. 2001. Technological Innovation in the Pharmaceutical Industry: The Use of Organizational Control in Managing Research and Development. Organization Science, 12(1): 19-36. • Chandrasekaran, A., Linderman, K. & Schroeder, R. 2012. Antecedents to Ambidexterity

Competency in High Technology Organizations. Journal of Operations Management, 30(1/2): 134-151. • Chang, Y., Hughes, M. & Hotho, S. 2011. Internal and External Antecedents of SMEs’ Innovation

Ambidexterity Outcomes. Management Decision, 49(10): 1658-1676.

• Chang, Y. & Hughes, M. 2012. Drivers of Innovation Ambidexterity in Small- to Medium-sized Firms. European Management Journal, 30(1): 1-17.

• Damanpour, F. 1996. Organizational Complexity and Innovation: Developing and Testing Multiple Contingency Models. Management Science, 42(5): 693-716.

25

• Dewar, R.D. & Dutton, J.E. 1986. The Adoption of Radical and Incremental Innovations: An Empirical Analysis. Management Science, 32(11): 1422-1433.

• Duncan, R.B. 1976. The Ambidextrous organization: Designing Dual Structures for Innovation, in R. Kilman & L. Pondy (eds.) The Management of Organizational Design. New York: North Holland: 167-188.

• Ettlie, J.E., Bridges, W.P. & O’Keefe, R.D. 1984. Organization Strategy and Structural Differences for Radical versus Incremental Innovation. Management Science, 30(6): 682-695.

• Germain, R. 1996. The Role of Context and Structure in Radical and Incremental Logistics Innovation Adoption. Journal of Business Research, 35(2): 117-127.

• Gerwin, D. & Barrowman, N.J. 2002. An Evaluation of Research on Integrated Product Development. Management Science, 48(7): 938-953.

• Gibson, C.B. & Birskinshaw, J. 2004. The Antecedents, Consequences, and Mediating Role of Organizational Ambidexterity. Academy of Management Journal, 47(2): 209-226.

• Greve, H.R. 2007. Exploration and Exploitation in Product Innovation. Industrial & Corporate Change, 16(5): 945-975.

• Grover, V. & Goslar, M.D. 1993. The Initiation, Adoption, and Implementation of Telecommunications Technologies in U.S. Organizations. Journal of Management Information Systems, 10(1): 141-163.

• He Z., & Wong, P. 2004. Exploration vs. Exploitation: An Empirical Test of the Ambidexterity Hypothesis. Organization Science, 15(4): 481-494.

• Hunter, J.E. and Schmidt, F.L. 1990. Methods of Meta-Analysis: Correcting Error and Bias in Research

Findings. Newbury Park, CA: Sage.

• Hunter, J.E. and Schmidt, F.L. 2004. Methods of Meta-Analysis: Correcting Error and Bias in Research

Findings, 2nd _{edition. Thousand Oaks, CA: Sage.}

• Ireland, D.R. & Webb, J.W. 2007. Strategic Entrepreneurship: Creating Competitive Advantage through Streams of Innovation. Business Horizons, 50(1): 49-59.

• Jansen, J.J.P., Bosch, F.A.J. van den & Volberda, H.W. 2005. Exploratory Innovation, Exploitative Innovation, and Ambidexterity: The Impact of Environmental and Organizational Antecedents.

Schmalenbach Business Review, 57(4): 351-363.

• Jansen, J.J.P., Bosch, F.A.J. van den & Volberda, H.W. 2006. Exploratory Innovation, Exploitative Innovation, and Performance: Effects of Organizational Antecedents and Environmental Moderators. Management Science, 52(11): 1661-1674.

26

• Kimberly, J.R. & Evanisko, M.J. 1981. Organizational Innovation: The Influence of Individual^ Organizational^ and Contextual Factors on Hospital Adoption of Technological and Administrative Innovations. Academy of Management Journal, 24(4): 689-713.

• Koberg, C.S., Detienne, D.R. & Heppard, K.A. 2003. An empirical test of environmental, organizational, and process factors affecting incremental and radical innovation. Journal of High

Technology Management Research, 14(1): 21-45.

• Levinthal, D.A. & March, J.G. 1993. The Myopia of Learning. Strategic Management Journal, 14: 95-112. • Li, C., Lin, C. & Chu, C. 2008. The Nature of Market Orientation and the Ambidexterity of

Innovations. Management Decision, 46(7): 1002-1026.

• Lubatkin M.H., Simsek, Z., Ling, Y. & Veiga, J.F. 2006. Ambidexterity and Performance in Small- to Medium-Sized Firms: the Pivotal Role of Top Management Behavioral Integration. Journal of

Management, 32(5): 646-673.

• March, J.G. 1991. Exploration and Exploitation in Organizational Learning. Organization Science, 2(1): 71-87.

• McDermott, C.M. & Prajogo, D.I. 2012. Service Innovation and Performance in SMEs. International

Journal of Operations & Productions Management, 32(2): 216-237.

• Menguc, B. & Auh, S. 2010. Development and Return on Execution of Product Innovation Capabilities: the Role of Organizational Structure. Industrial Marketing Management, 39(5): 820-831. • Mom, T.J.M, Bosch, F.A.J. van den & Volberda, H.W. 2007. Investigating Managers’ Exploration

and Exploitation Activities: the Influence of Top-down, Bottom-up, and Horizontal Knowledge Inflows. Journal of Management Studies, 44(6): 910-931.

• Morgan, R.E. & Berthon, P. 2008. Market Orientation, Generative Learning, Innovation Strategy, and Business Performance Inter-Relationships in Bioscience Firms. Journal of Management Studies, 45(8): 1329-1353.

• Narver, J.C., Slater, S.F. & MacLachlan, D.L. 2004. Responsive and Proactive Market Orientation and New-Product Success. The Journal of Product Innovation Management, 21(3): 334-347.

• O’Reilly, C.A. & Tushman, M.L. 2011. Organizational Ambidexterity in Action: How Managers Explore and Exploit. California Management Review, 53(4): 5-22.

• Ouchi, W.G. 1977. The Relationship between Organizational Structure and Organizational Control.

Administrative Science Quarterly, 22(1): 95-113.

• Pierce, J.L. & Delbecq, A.L. 1977. Organization Structure, Individual Attitudes and Innovation.