Are large MNEs the better innovators?

Are large MNEs the better innovators?

An empirical study of European MNEs on organizational complexity,

innovation and the moderating role of board characteristics

Niklas Blohm

S3488225

E-Mail address: n.blohm@student.rug.nl

Course: Master’s Thesis IB&M

Course code: EBM719A20.2017-2018.2

Master Thesis

MSc International Business & Management

18

_{June 2018}

Supervisor: dr. Olof Lindahl

Co-Assessor: dr. Sathyajit Gubbi

I

ABSTRACT

Over the past decades, large multinational enterprises (MNEs) became even larger, entered more markets and operated in more countries. While constantly growing, MNEs are also becoming more and more complex. Does this complexity have an adverse effect on their ability to innovate? Prior studies investigating the impact of organizational complexity on the innovation performance of MNEs came up with ambiguous results. Arguing from an attention-based view perspective and in consideration of the social identity theory, this study was formulated to add valuable insights to previous findings by assessing the importance of board characteristics in moderating the complexity-innovation relation. Additionally, a new measurement of structural complexity introduces an international perspective to the organizational complexity literature that captures the innovation power of the MNE subsidiary network. Using a negative binomial regression based on data on the organizational structure and board composition of 130 European MNEs, operating in 30 patent-intensive industries, this study produced interesting results, which partly differ from the predictions. The findings find support for the hypothesis that larger MNEs are more innovative. Additionally, a larger and nationally diverse board further promotes the positive effect of firm size on innovation. In contrast with the predictions made, no support was found for any effects of an MNE’s subsidiary structure on innovation performance.

Keywords: MNE, organizational complexity, innovation, attention-based view, board of

II

TABLE OF CONTENT

INTRODUCTION ... 1

THEORY AND HYPOTHESES ... 3

Innovation ... 3

Organizational complexity ... 5

Introducing the board characteristics ... 6

Organizational complexity & innovation ... 7

The attention-based view ... 12

The influence of board characteristics ... 13

National board diversity ... 13

Board size ... 16

Conceptual model ... 17

METHODOLOGY ... 18

Sample and data collection ... 18

Variables ... 20 Dependent variable ... 20 Independent variables ... 21 Moderator variables ... 22 Control variables ... 22 Research method ... 23 Robustness checks ... 25 RESULTS ... 26 Descriptive statistics ... 26 Correlation ... 27 Regression results ... 28

Robustness check results ... 31

DISCUSSION ... 33

CONCLUSION ... 36

Key findings ... 36

Contributions and implications ... 36

Theoretical implications ... 36

Managerial implications ... 37

Limitations and future research ... 38

III

APPENDICES ... 48

Appendix I: Patent-intensive industries ... 48

Appendix II: List of companies ... 49

Appendix III: List of variables ... 51

Appendix IV: Normality and homoskedasticity test ... 52

Appendix V: Distribution of the dependent variable ... 54

Appendix VI: Variance Inflation Factor tests ... 55

Appendix VII: Marginal effects of interaction terms ... 56

IV

FIGURES

Figure 1: Conceptual Model

Figure 2: Histogram distribution of residuals Figure 3: Residual vs. fitted plot

Figure 4: Histogram dependent variable

18 52 52 54

TABLES

Table 1: Means, standard deviations, median, minimum and maximum values Table 2: Correlation Matrix

Table 3: Negative binomial regression predicting Patents Table 4: List of patent-intensive industries

Table 5: List of variables

Table 6: Skewness / Kurtosis tests for Normality

Table 7: Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Table 8: White’s test for heteroskedasticity

Table 9: Dependent variable, detailed descriptive statistics Table 10: VIF incl. R&D and Subsidiaries

Table 11: VIF excl. Subsidiaries

Table 12: VIF excl. R&D Expenditures Table 13: NB regression excl. country Table 14: NB regression incl. ROE

Table 15: NB regression incl. Foreign_Ratio as independent variable

Table 16: Means, standard deviations, median, minimum and maximum values (3-year time lag)

Table 17: Correlation Matrix (3-year time lag) Table 18: NB regression incl. time lags

Table 19: NB regression incl. standardized predictors

V

LIST OF ABBREVIATIONS

ABV Attention-based view FDI Foreign direct investment IMF International Monetary Fund MNE Multinational enterprise NB negative binomial

OECD Organisation for Economic Co-operation and Development OFDI Outward foreign direct investment

OLS Ordinary least squares R&D Research and development ROA Return on assets

ROE Return on equity SIT Social identity theory

1

INTRODUCTION

Regarding the past decades, it is striking that large multinational enterprises (MNEs) became even larger and increased the number of countries they operate in (Reiche et al., 2015). European MNEs play a central role in this development since five of the Top 10 countries in terms of outward foreign direct investment (OFDI) are the Netherlands, France, Ireland, Spain, and Germany, together exceeding the leading country in terms of OFDI (the United States) by more than 50 billion US$ (UNCTAD, 2017). While enlarging their market and gaining more resources, growing MNEs face the risk of higher complexity. Does this complexity have an adverse effect on their ability to innovate? This question is crucial since innovation is essential for MNEs and a key for the success and the survival of firms in highly competitive markets (Bartlett & Ghoshal, 1989; Cefis & Marsili, 2006).

In this thesis, innovation is defined as the successful production and assimilation of novelty (Parajón Collada, 1999; Acemoglu et al., 2011). Organizational complexity can be understood in many ways, with especially firm size and structural complexity having a rather strong impact on innovation (Damanpour, 1996). The latter one is determined e.g. by the number of locations, different jobs or hierarchical ranks (Mileti et al., 1977). However, so far, theoretical approaches and studies on this topic, couldn’t present a clear picture of the relationship between organizational complexity and innovation. On the one hand, researchers propose that complexity positively influences innovation performance due to a more diversified knowledge base, a greater variety of specialists and more product development experience (e.g. Young et al., 1981; Damanpour, 1996), while others have shown that high structural complexity determines the processing of radical innovations (Ettlie et al., 1984). On the other hand, some authors argue for a negative effect due to more formalized structures, more standardized behavior of managers and organizational inertia (e.g. Aldrich & Auster, 1986; Hitt et al., 1990) and show that larger firms are slower in responding to new opportunities (Cohen & Klepper, 1996).These contradicting results might be traced back to conceptual or methodological factors like a missing distinction between firms that adopt innovations and firms that generate innovations (Damanpour & Wischnevsky, 2006), or differences in complexity measurements. In order to move forward in the research about organizational complexity and innovation, it is of critical importance to understand the applied concepts, methods, and factors influencing the relationship.

2 organizational structure on innovation has been investigated. Chang & Harrington (1998) examined the role of centralization of the organization on innovation performance, whereas Song & Chen (2014) looked at organizational attributes and the roles of strategic planning. While moderating factors regarding the influence of organizational complexity mainly have been introduced on the country- or firm- level, only some had looked for them on the individual-level, like Krause (2009) who studied the impact of individuals on the control of organizational complexity.

Studies in this context, examining the effect on innovation performance, are not existent. Further research is necessary to show how individuals can influence the complexity-innovation relationship. Thus, this study investigates the impact of board characteristics on this relationship. Although prior research acknowledges that board characteristics can affect innovation performance, research in the context of organizational complexity still needs to be done. Furthermore, contradictory theories predict different effects of board characteristics. Whereas the resource dependency theory (Pfeffer & Salancik, 1978) and upper echelon theory (Hambrick & Mason, 1984) suggest that a more divergent board lead to better performance, the original version of the social identity theory (Tajfel, 1978) predicts a negative impact. However, to advance the research on organizational complexity and innovation this thesis introduces national board diversity and number of board directors as moderating factors since they haven’t been examined in this context before, to the best of my knowledge. I predict for both moderators to positively influence the complexity-innovation relationship. To address the research gap, this thesis is based on the following research question:

“What effect does organizational complexity have on the innovation performance of Europe based MNEs? How is this relationship affected by board characteristics?”

3 multiple countries and industries. Using a company´s subsidiary networks an indicator of structural complexity, including a distinction of emerging market and developed market subsidiaries, presents a novelty since prior studies mainly focused on number of role specializations, departments involved in the decision-making process, and different products (e.g. Aiken et al, 1980; Häggman, 2009; Zhou & Wan, 2017). The thesis also introduces new theoretical approaches to the topic with the attention-based view, to show how complexity can influence a firm’s strategic orientation, and the social identity theory, to examine the effects of a national diverse board on the complexity-innovation relationship. This new perspective requires a measurement incorporating national differences in order to observe how the spread of innovation within an MNE network is influenced.

Due to a new measurement and new theoretical approaches, the findings of this thesis will provide new knowledge regarding the topic. Additionally, analyzing the impact of board characteristics on the complexity-innovation relation helps in filling the research gap regarding individual influencing factors.

In order to test the portrayed relationships, a negative binomial regression is performed based on archival data. Furthermore, additional tests are performed checking for robustness of the method used, revealing that the main results hold true even after changing indicators.

This thesis consists of six parts. The introduction presents the aim and the relevance of this thesis, while the theory section gives key definitions of the relevant topics, presents prior research as well as the hypotheses and the theoretical model. The methodology part presents the research design of this study, including the research model, data, and sample, as well as all the variables and robustness checks. Empirical results are shown in the analysis section. In the final two sections, the results from the analysis are discussed and a conclusion of this thesis is drawn, including future outlook and theoretical as well as managerial implications.

THEORY AND HYPOTHESES

Innovation

Innovation exists in a variety of fields and areas from technology to business, thus having various definitions (Baregheh, Rowley & Sambrook, 2009). It is a broad term that includes several kinds of innovation, like business model, process and product innovation (Khazanchi et al., 2007).

4 (Damanpuor, 1996). Innovation is key for the business success and survival of firms in highly competitive markets and fast-paced industries (Cefis & Marsili, 2006; Anderson et al., 2014). Innovation positively influences firm performance and is an important aspect of a firm’s comparative advantage (Salomo et al., 2007; Löfsten, 2014). Therefore, constant innovation is crucial for MNEs since they compete in globally competitive markets.

Although used in some studies as an indicator for firm-level innovation (Raymond & St. Pierre, 2010), R&D activities capture innovation input but not necessarily innovation output or other innovation-related investments in technology or employee training (Hall, 2010). Furthermore, R&D expenditures are more related to process innovation, hence protected by hiding it from the public (Cohen & Klepper, 1996). In order to be able to measure innovation performance, this study is regarding industries where the innovation output is captured in form of patents. Thus, innovation in this thesis is defined as the successful designing, producing and implementing of new products and technology, offering new solutions to problems (Nakata & Sivakumar, 1996; Parajón Collada, 1999).

One important component leading to firm innovation is the company’s employees. The more they are involved, the more innovative is a firm. For an ethnically diverse workforce, this effect is even stronger, leading to superior innovation performance. Although deemphasized by some researchers (Joshi & Roh, 2009), various studies underline that companies benefit from differences in values, perspectives, and cultures due to new sources of knowledge and new perspectives. Hence, the number of thought categories and mental images within a firm grows and the chance for novel ideas to occur increases, resulting in more innovation (Kashima et al., 1995; Konrad et al., 2000; Yang & Konrad, 2011; Mohammadi et al., 2017). A group-level study among Fortune 500 companies (Cady & Valentine, 1999), as well as individual-level research by Leung et al. (2008), underline that racial diversity and access to information from several cultures lead to more innovation.

Consequently, external innovativeness is also an important factor influencing innovation performance of an MNE. External pressures from the environment stimulate firms to innovate when trying to keep up with the innovative competition. Additionally, spillover effects from the local stakeholder network can occur, representing one of the main sources for innovation and leading to an enhancement of a firm’s innovation performance (Powell et al., 1996; Li et al., 2018).

5 Miller, 2008). Knowledge heterogeneity between different countries and geographical diversity along a firm’s locations positively influences innovation (Hargadon & Sutton, 1997). In the past years, emerging markets gained importance as innovation source, in particular for reverse innovation that is first adopted in poor countries and later transmitted to developed countries. This development is highly relevant for the strategic management of MNEs (Govindarajan & Ramamurti, 2011).

According to Un (2016), solely domestic firms are less innovative than MNEs because of a limited ability to identify and transfer diversified knowledge that is key for product innovation. Goerzen & Beamish (2005) on the other hand assume that diversity, as seen in MNE networks, might lead to difficulties in exchanging information due to differences in language, vocabulary, and objectives, resulting in higher communication problems and reduced social integration both being an obstacle for organizational innovation. Whether diversity of large MNEs is enhancing or hindering innovation will be discussed in one of the following sections by regarding the relation between organizational complexity and innovation.

Organizational complexity

The concept of organizational complexity consists of various parts and manifests itself in several ways. In general, organizations must cope with internal and external complexity. Studying the impact of organizational complexity on innovation performance, this thesis has its focus on internal complexity only. There are three types of processes that determine internal complexity: materials transformation, information processing and communication with stakeholders (Dervitsiotis, 2015). The extent of organizational complexity can be defined by the amount of differentiation or variety that exists within different elements of the organization e.g. in product lines, markets or geographical locations. This complexity affects the sense-making of organizational members’ current perceptions (Dooley, 2002). Due to the description above companies can be defined as complex systems with the function to transfer information and materials between tasks (Baldwin, 2008). The costs accompanying the coordination of such complex tasks, however, might reduce the organizations´ benefits from the diversification into affiliated businesses, like new product varieties or access to new knowledge (Zhou, 2011; Zhou & Wan, 2017).

6 resources, and more experience while being more formalized and bureaucratic (Nord & Tucker, 1987; Hitt et al., 1990; Berends et al., 2014). Small firms, on the other hand, are more flexible than large firms, but they lack resources and various capabilities like organizational and marketing skills (Berends et al., 2014).

Structural complexity in an organization is determined by the number of different products, different locations, jobs performed, hierarchical ranks and the degree of the centralization of decision making (Mileti et al., 1977; Ettlie et al., 1984; Chang & Harrington, 1998). The more differentiation along spatial, hierarchical and functional dimensions exist, the more complex is an organization’s structure.

Introducing the board characteristics

The board of directors is a group consisting of elected individuals that represent the shareholders. It is mandatory to have a board for every public company. While being the most influential actors regarding strategic direction, and decision-making, boards develop the corporate strategy, introduce new strategic changes and goals and implement them within the company (Finkelstein & Hambrick, 1996). The board of directors takes the role of hiring, compensating and monitoring the top management on behalf of the shareholders, responding to takeover threats, and providing the company with resources (Finkelstein & Hambrick, 1996; Corbetta & Salvato, 2004). It is therefore responsible for governing the firm and for overseeing all relevant strategic decisions including the alignment of attention to the strategic orientation (Cannella et al., 2015). Additionally, the board is the connection between the company and external organizations and therefore responsible for gathering information from external partners (Pfeffer & Salancik, 1978). Given its role in the company, the board of directors can significantly influence the extent of strategical orientation towards innovation by fostering a context that supports executives´ striving for innovation (Zona et al., 2013).

According to theory (e.g. upper echelon theory, social identity theory), and previous studies, board characteristics, like composition, size or diversity, influence a firm´s behavior, performance, and innovation (Tajfel, 1978; Hambrick & Mason, 1984; Anderson et al., 2004; Guest, 2009). Board characteristics, however, vary among different national cultures and company types (Corbetta & Salvato, 2004). This study focuses on two board characteristics in order to examine their effect on the complexity-innovation relationship.

7 theory, a diverse board in terms of nationality should increase the firm’s access to external networks, diversified knowledge, and information and help the firm to get familiar with a foreign market (Carter et al., 2010). The upper echelon theory proposes that the organization reflects the management team characteristics differing in psychological and demographic attributes (Hambrick & Mason, 1984). The more diverse the team’s demographic attributes, the more diverse its cognitive attributes which positively affects firm performance. However, there are also theories that seem to be contradicting at first sight. The social identity theory (SIT), for example, proposes that board members form social groups based on demographic attributes, favoring in-group members over out-group members, thus impeding the communication among the board members (Tajfel, 1978).

A second characteristic introduced in this thesis is board size. It is an important parameter regarding the monitoring, controlling and advising of a company and its executives (Andrés et al., 2005; Frick & Bermig, 2009; Ahern & Dittman, 2012). Referring to the total number of board members, the literature shows that the board size affects the company in several ways. A larger board size is associated in the literature with problems of communication, coordination, and control (Smith et al., 1994) but also with improved creativity and performance (Webber & Donahue, 2001).

In the following, the theoretical concepts will be conflated, and the relationships between complexity, innovation, and the board will be discussed.

Organizational complexity & innovation

8 factor influencing the complexity-innovation relationship might be the difference in the measurement of organizational complexity in prior studies.

Basically, two forms of organizational complexity, organizational size, and structural complexity have an impact on innovation, especially on the implementation of product innovation (Damanpour, 1996). As stated by Blau (1970), size is an important factor influencing the structure and the processes of an organization. The impact of the firm size on innovation performance, though, is a topic in need of further discussion.

On the one hand, large firms have more financial resources, more product development experience and are rather able to handle potential losses resulting from unsuccessful innovations, resulting in more innovation projects undertaken by larger firms (Damanpour, 1996; Damanpour & Wischnevsky, 2006; Berends et al., 2014). Furthermore, their technological capabilities are often superior compared to small firms. Given the importance of technological capabilities for innovation, large firms are more likely to become aware of scientific breakthroughs and innovative ideas at an earlier stage while pursuing the most promising ones (Chandy & Tellis, 2000). The more complex and diversified facilities of large organizations, economies of scale and greater market size promote the development of innovation strategies,as well as the recognition and adoption of innovative ideas and give them an advantage over smaller companies (Nord & Tucker, 1987; Damanpour & Wischnevsky, 2006). In contrast to large firms, smaller firms with limited resources and less experience face huge challenges in their innovation efforts (Berends et al., 2014).

On the other hand, some authors argued that largeness not always results in greater innovativeness (Hage, 1980). Researchers found large firms to be more formalized, with higher levels of inertia, less managerial commitment to innovation, and more bureaucratic structures which can negatively affect innovation, creativity and the response to opportunities (Pierce & Delbecq, 1977; Hitt et al., 1990; Cohen & Klepper, 1996). Löfsten (2014) even shows that innovation performance of Swedish medium-sized technology firms is not affected by firm size at all. Hence, prior results are contradicting.

9 ideas can be realized, and the innovation output will be higher. Supporting this, Chandy & Tellis (2000) showed that recently large firms are more likely to introduce radical innovation in contrast to smaller firms, especially outside the US. Hence, the belief of large firms suffering from inertia and being unlikely to introduce radical innovations seem to be refuted.

Additionally, the more employees the larger the pool of innovation sources. Especially in large MNEs with a culturally diverse workforce, including more people and more nationalities, exist a high potential for a variety of innovative ideas since innovation spurs from countries different from the home country. Small firms might be able to communicate their ideas faster, but the number of ideas should be lower than in large, multinational organizations, leading to less innovation output. In support of this view, Damanpour (1992) shows a positive and statistically significant relationship between organizational size and innovation, with a stronger effect in manufacturing and profit-making organizations compared to service and non-profit making companies. Since this thesis focuses on patent-intensive industries, dominated by manufacturing industries, I come up with the following hypothesis:

H1: The greater the size of an MNE, the higher its innovation performance.

Research on structural complexity in an organization, determined by the differentiation along spatial, hierarchical and functional dimensions (Mileti et al., 1977) seems to reveal a more explicit image regarding its influence on innovation. Although high product variety can worsen firm performance (Zhou & Wan, 2017), the impact of a high degree of structural complexity on innovation was found to be positive and significant in most of prior research, e.g. because of a more diversified and deeper knowledge base and a greater variety of specialists (Damanpour, 1991; Damanpour, 1996). The relationship between complexity, in terms of a large subsidiary network, and innovation, however, still needs to be tested.

10 In this thesis, structural complexity is determined by the amount of variety in geographical locations. Although the spatial complexity associated with geographic dispersion of personnel might impede the information processing (Sanders & Carpenter, 1998), I argue, that there is more innovative potential within a structural complex MNE that is geographically diversified and has subsidiaries in several host countries. As shown in previous sections, innovation is higher with a more ethnically diverse workforce (Mohammadi et al., 2017). Also, knowledge heterogeneity between different countries and the ability of MNEs to better identify and transfer this diversified knowledge are positively influencing innovation (Hargadon & Sutton, 1997; Un, 2016).

Concluding the above, a highly diversified company is assumed to be more innovative, resulting in the following hypothesis:

H2: The higher the structural complexity of an MNE, the higher its innovation performance.

Additionally, in this thesis, it is assumed that not every location contributes in the same manner to the complexity of the subsidiary network. Supposing that there is a difference between subsidiaries from emerging countries and those from developed countries due to a different economic background (London & Hart, 2004), the ratio of emerging market subsidiaries should have an impact on the positive relationship between structural complexity and innovation performance.

In the past years, emerging markets rapidly grew and developed economically (Marquis & Raynard, 2015) leading to a change regarding the locus of innovation in the global economy. Innovations from emerging markets, however, appear to be rare (Govindarajan & Ramamurti, 2011). This might be caused by the fact that emerging market MNEs have weaker innovation capabilities compared to MNEs from developed markets (Luo & Tung, 2007). Thus, MNEs from developed markets with superior innovation capabilities might rather profit from the enormous potential that lies in emerging markets and use it to enhance their innovation performance.

11 bottom of the pyramid customers lead to the requirement of innovative ideas and hence to a rise of innovative products first adopted in emerging markets and then tickling up to developed markets (London & Hart, 2004). MNEs investing in emerging markets thus can profit from FDI spillovers due to learning from local competitors, suppliers, and customers about new technologies (Govindarajan & Ramamurti, 2011). In combination with external pressures, these spillover effects lead to innovation (Li et al., 2018).

To be able to profit from spillover effects, a company needs to be embedded in the local context. Otherwise, this might hamper the spur of innovative ideas. Another hindrance might occur when developed market MNEs are not aware of learning opportunities in local markets or are not motivated to learn. Also, internal resistance to transferring radical innovations might be problematic if there is a risk of cannibalization regarding existing products in the developed countries (Govindarajan & Ramamurti, 2011).

Nonetheless, if knowledge differences are larger between developed and emerging markets than between two developed markets and taking the strategical relevance of emerging markets as an important source of innovation into account (Govindarajan & Ramamurti, 2011), I argue that having more subsidiaries from emerging markets is beneficial for an MNE. Additionally, the more subsidiaries an MNE has in emerging markets, the more it is supposed to be aware of the importance of these markets. Therefore, I propose that MNEs with a strong presence in emerging markets are motivated to learn and aware of learning possibilities. Furthermore, financial resources and innovation capabilities of European MNEs allow them to cope with innovative ideas from emerging markets. Thus, an extension of the second hypothesis is made by adding the emerging market ratio as a moderating factor:

H3: The higher the ratio of emerging market subsidiaries to the number of all subsidiaries, the higher is the positive effect of a high structural complexity on innovation performance.

12

The attention-based view

One of the strongest arguments for structural complexity to have a positive impact on innovation is that a complex organization has access to diverse knowledge sources from its network. But what if the most innovative knowledge resides in markets the MNE does not pay attention to?

The ABV assumes that an organization is consisting of complex networks with its strategic orientation defined by cognitive and social structures through which information is transmitted (Ocasio, 1997). Those channels influence the attention decisions made in an organization concerning different environmental issues. Ocasio (1997) defines attention as noticing, encoding, interpreting and focusing on particular issues and available solutions. In a later work (2011), he identified three types of attention. The first is the attention perspective on how to allocate firm resources equivalent to the strategic orientation.

Second, the attention engagement, defined by the process of intentional allocation of cognitive resources to guide strategic actions like problem-solving, planning or sense-making. Balancing between committing resources to identify business ideas from specific subsidiaries and diversifying attention by focusing on multiple subsidiaries is important to gain novel business ideas (Rerup, 2009).

Third, attention selection is determined by the alignment of attention engagement and perspective. Decision-makers must be selective concerning the issues they tackle because attention is a scarce resource. By tackling certain issues at one time, they often ignore other issues in the environment (Haq et al., 2017). If attention engagement and perspective are not in line, attention dissonance occurs, leading to a confusion at the subsidiary level (Ocasio, 2011). This is problematic since subsidiaries, due to their dual embeddedness in the local context and the organizational network, are an important source of new business and innovation ideas (Strutzenberger & Ambos, 2014).

13 The selective attention focus of decision makers is influenced by bounded rationality and managerial cognition (Ocasio, 1997). Managerial attention is in its nature limited wherefore some strategically relevant environmental signals are not noted at all (Shepherd et al., 2017). According to the ABV, the higher the number and the diversity of a company’s subsidiaries, the less is the chance that an innovative idea from a subsidiary far away from the headquarter can reach it although the potential for innovation from distinct markets is enormous. Reasonable explanations are that attention is a scarce resource and that managers are more reluctant vis-à-vis alien-market subsidiaries compared to subsidiaries from markets they are familiar with.

This section showed that, although organizational complexity is assumed to have a positive effect on innovation performance, not all innovative ideas might be captured, and that more innovative potential resides within an MNE than actually exploited. Hence, there might be a moderating factor in facilitating the innovation flow within an MNE and broadening its attention engagement. One factor that was shown to have an impact on innovation performance but was barely looked on in terms of organizational complexity is board characteristics (Carter et al., 2010). Thus, the following section will introduce national board diversity and board size as possible moderating factors and discuss their influence on the complexity-innovation relation in MNEs.

The influence of board characteristics

This thesis introduces two board characteristics that have an impact on the complexity-innovation relationship: national board diversity and board size.

National board diversity

In the literature, there are various theories portraying the impact of board diversity in general on firm performance. In this thesis, I focus on the social identity theory (SIT) in order to explain the relation of national board diversity to the complexity-innovation relationship in MNEs. This focus is chosen because of the theoretical background regarding the ABV. The SIT, in this context, explains best the distribution of attention.

14 communication behavior between directors with different demographic attributes and diminishing the proposed benefits of diversity like a wider range of perspectives (Tajfel & Turner, 1986; Ali et al., 2014).

As pointed out, social identification occurs due to shared demographic attributes like gender, nationality but also language. If the headquarter and subsidiary share the same language there is a better understanding of knowledge and it is more likely that a common social identity will develop with a positive effect on tacit knowledge flows, e.g. innovative ideas (Reiche et al., 2015). This is relevant when examining the role of social identity in MNEs.

So far, SIT is a well-established concept in social psychology, but it is new to the context of MNEs. Recent studies in the MNE context are related to the consequences of the identities that subsidiaries hold vis-à-vis the MNE (Vora & Kostova, 2007), the effect of intra-group cooperation (e.g. Hinds & Mortensen, 2005) and the inter-unit interaction and cooperation in MNEs (Reiche et al., 2009). While some scholars have found negative implications of having a shared social identity in form of resistance to organizational change (Bouchikhi & Kimberly, 2003) most of the studies on social identity and MNEs found that a shared identity among headquarter and subsidiaries is beneficial regarding knowledge exchange (Reiche et al., 2015). The subsidiary’s role in knowledge transmission depends on its resources and the overall MNE strategy and attention perspective (Reiche et al., 2015).

The question that arises is if the positive impacts as discussed above outweigh the negative impacts a national diverse board might have according to the SIT. While some researchers propose that a diverse board leads to a better performance due to a greater knowledge base, more creativity as well as a higher strategic decision-making quality (Erhardt et al., 2003), others argue that higher diversity leads to greater time and effort necessary to find consensus, therefore having a negative impact on performance (Knight et al., 1999).

15 more varied strategical orientation and thus an increased contribution of the subsidiaries in terms of new knowledge. By enabling the release of the innovation potential that resides in distant markets, innovation performance is enhanced. A homogenous board with limited focus due to one strong social group might miss substantial ideas because of its lack of attention paid to subsidiaries perceived as an out-group. It ratherrelies on ideas coming from similar cultural backgrounds, probably not essentially different from their own ideas. Thus, a homogenous board of an MNE with high structural complexity, i.e. more locations, might lead to even more unrecognized sources of innovative ideas. On the contrary, a highly diversified board with a less strong in-group better distributes its attention and would rather recognize ideas and therefore profit from the higher amount of potential innovation sources. This results in the following hypothesis:

H4a: The more diverse the board of the firm is in terms of nationality, the higher is the positive effect of a high structural complexity on innovation performance.

Additionally, firms with a more diverse board in terms of nationalities tend to have broader cognitive maps and a higher number of thought categories but also a greater pool of knowledge and perspectives fostering creativity, alternative solutions, and innovation (Yang & Konrad, 2011; Zona et al., 2013). Therefore, it is not only assumed that innovations and ideas from host countries are more recognized in a firm with a national diverse board but innovations and ideas in general that arise in a large MNE. Broader cognitive maps and a higher number of thought categories improve the board´s capability to recognize, process and link new innovative ideas and help to overcome a selective attention focus that might be influenced by bounded rationality or limited managerial cognition (Maitland & Sammartino, 2015). Therefore, national board diversity has a positive impact on a large MNE´s innovation performance even if it is not present in many foreign locations. This assumption leads to the following hypothesis:

16 Board size

Another board characteristic influencing the relation between organizational complexity and innovation performance is the board size. It is as an important factor regarding monitoring and control activities as well as the advising of a company and its executives (Andrés et al., 2005; Frick & Bermig, 2009; Ahern & Dittman, 2012). More members might improve the board’s competence but lower the overall level of communication, coordination and social integration (Smith et al., 1994). This results in a more complicated decision-making process with the final decision being a compromise of all individual opinions. The larger the group, the harder it is to find a compromise, especially when dealing with the complexity and riskiness of innovation projects (Zona et al., 2013). Another reason for the negative impact of larger boards might be the perceived ability to contribute less to decision-making resulting in a lack of motivation to contribute at all (Carpenter & Westphal, 2001). As found by Guest (2009) in his study on publicly listed UK firms, the problems leading to a negative relation between board size and firm performance, like poor communication and decision-making, occur in the advisory rather than the monitoring role of the board. Although prior studies found evidence for a negative correlation between too large boards and financial performance (Yermack et al., 1996; Bhagat & Black, 1999), other researchers showed that this inverse correlation is not robust to the choice of performance measure (Bhagat & Black, 1998; 1999).

Nevertheless, several other researchers also emphasize the positive effects of a large board. In general, the ability to handle complexity is superior for teams compared to individuals due to more abundant skills and greater problem-solving as well as information-processing capacities (Sanders & Carpenter, 1998). According to Hambrick and D’Aveni (1992), the number of team members determines the availability of its capacities and resources. Therefore, larger boards have a greater pool of knowledge and collective information (Guest, 2009; Jackling & Johl, 2009). This abundance of cognitive resources, in turn, might lead to improved group decision-making quality, creativity and performance by providing expert advice and access to critical information and resources (Webber & Donahue, 2001; Guest, 2009).

17 its attention stock. Arguing from an ABV perspective, a higher stock of attention enables the board to distribute the attention among a higher number of subsidiaries and employees. Regardless of the different nationalities, a larger board has more resources usable for attention engagement. Therefore, a large board guarantees even large MNEs to still pay enough attention to every subsidiary. These assumptions lead to the following hypothesis:

H5a: The larger the board, the higher is the positive effect of large organizational size on innovation performance.

While the influence of board size on the relationship between firm size and innovation performance is assumed to be direct and positive, its impact on the relationship of structural complexity and innovation performance is rather indirect. The size of the board is said to affect the degree of diversity within a group (Blau, 1977). Therefore, larger boards are more likely to feature dissimilarity and diversity among its board members (Wiersema & Bantel, 1992; Goodstein et al., 1994). Consequently, the relationship suggested in Hypothesis 4a should also hold in the case of a larger board. Additionally, Sanders & Carpenter (1998) propose that a larger board increases the information processing capacity of the group and thus enhance the capability to cope with international complexity. Since structural complexity in this thesis is determined by the international network of an MNE, i.e. its international complexity, I come up with the following hypothesis:

H5b: The larger the board, the higher is the positive effect of a high structural complexity on innovation performance.

Conceptual model

18 Figure 1: Conceptual Model

METHODOLOGY

Sample and data collection

The sample of this study consists of 130 European MNEs, including Dutch, French, Irish, Spanish and German companies. The decision for European MNEs from those five countries has several reasons. First, the selected countries are the European countries with the largest amount of OFDI (UNCTAD, 2017). Additionally, European firms face relatively small domestic markets meaning that there is a larger concentration of international firms (Knight et al., 2004). Smaller domestic markets result in investments abroad and therefore create a large testing ground for the hypotheses of this study.

Furthermore, one criterium in favor for European firms is the availability of secondary data in archival databases.

19 This thesis defines MNEs as companies, having subsidiaries in at least one country other than the home country. Filtering for firms with this requirement, and all other requirements discussed below can be done via the Orbis database. In order to have access to the data needed for the study, this thesis focuses on publicly listed firms only. The chosen year for the data used in this study is 2016 for the patents since this is the last year with available information for all companies in the sample.

To determine the impact of the predictor variables on the patent output, this thesis uses a time lag for the main analysis. It has been recognized that innovation is a time-dependent process with a lag between having an idea and the implementation of it (Ettlie & Vellenga, 1979; Alegre et al., 2006). The exact lag is hard to define since it varies among industries and countries. While some researchers argue it takes a development time of ten years for innovations to breakthrough (McDermott & O’Connor, 2002) others found evidence for a shorter product development time, e.g. less than two years (Ali et al., 1995). Previous research on manufacturing companies showed no further effect for a time lag greater than seven years (Pakes & Griliches, 1980; Hall et al., 1984). As also recommended by the OECD, many studies examining innovation performance measured by patents use a time lag varying between two and four years (e.g. Ahuja & Katila, 2001; Alegre et al., 2006).

Although barely recognized in prior studies, the duration of the patent application process should be considered additional to the product development time. In Europe, this process can take up to seven years depending on the innovation itself, the country where the patent is announced and the industry, but with 2-3 years being the norm for manufacturing firms (Ernst, 2001; Bardehle Pagenberg, 2010).

Incorporating both factors, this study uses a 6-year time lag (three years development time followed by three years application process) to capture the effect of the predictor variables on innovation performance.

Since time lags are included, the company data must be also available for the years 2010 and 2013 (for robustness check with a different time lag). Hence, companies with missing financial statements for one of those years are excluded.

In order to be able to examine the impact of organizational complexity this thesis includes large and very large companies as defined by the Orbis database.1 The decision for large and very large companies is motivated by the fact that patenting behavior differs between large and

20 small firms (Hagedoorn & Cloodt, 2003). Therefore, studying both might influence the outcome due to unobservable factors.

Finally, I set industry filters to only include industries that are patent-intensive for the purpose of comparability in terms of development and use of patents. Included are the 30 most patent-intensive industries in 20162. After filtering I end up with 145 companies. Due to missing data issues, the final number of observations is 130. Given the rule of thumb for the number of observations required in regression models, the minimum number of observations in this study would be 90 (nine predictor variables x 10) (Harrell, 2015). Hence, the size of this sample is sufficient in order to produce meaningful results.

Variables

Dependent variable

In order to capture innovation performance, this study will look at granted patents. This is in line with prior studies (e.g. Ahuja & Katila, 2001). Although, there exist some critic on this measurement, e.g. because patents cannot capture incremental innovations, or because of differences in the patenting behavior between sectors, countries and firm size (Hagedoorn & Cloodt, 2003; Rodriguez-Pose & Di Cataldo, 2014) this thesis will use the number of patents as a proxy for innovation performance for several reasons. First, the composition of the sample helps to prevent some of the problems by incorporating only large firms, by focusing on similar industrial sectors and by regarding countries with a similar patent application process. Second, patents promote ex-ante innovation because they create monopoly rents ex-post, thus capturing the outcome of an innovation process (e.g. Grossman & Helpman, 1991; Scotchmer, 1999; Acemoglu et al., 2011). Third, they are a representative measure of technological novelty (Griliches, 1990). Fourth, Hagedoorn and Cloodt (2003) found that there is an overlap between any of the commonly used indicators for innovation performance (R&D expenditure, patent counts, patent citations, number of new products). The more an industry is characterized by high patenting intensity the higher is the overlap, hence any indicator could be used to measure innovation performance. Patent data is taken from Orbis.

21 Independent variables

In order to measure the impact of organizational complexity on innovation performance, this study includes two separate independent variables, (1) organizational size and (2) structural complexity.

In line with previous studies, organizational size is measured by the number of employees (Corvin 1975, Kim 1980, Damanpour 1996). In the literature, there also exist other measurements of organizational size like annual budget (Damanpour, 1987) or total assets (Gubbi, 2015). However, this study argues that innovation occurs on firm or individual level and that attention is a scarce resource, difficult to distribute equitably over the firm network. Imagine a company with relatively few employees but an enormous budget. Using a financial measurement method would lead to the assumption that the company is large and therefore organizational complex, although there would not be any problem of attention scarcity. Therefore, a personnel indicator is more suitable for this study than a financial indicator.

For structural complexity, previous research used measurements like units below the chief executive level, number of occupational specialties, different products or departments involved in a decision-making process (Aiken et al, 1980; Damanpour, 1996; Häggman, 2009; Zhou & Wan, 2017). However, none of these measurements captures national differences within a company network that are important for this study. The few papers that recognized the importance of a geographical indicator to determine structural complexity used a regional-focused and revenue-based measurement (e.g. Bushman et al., 2004). This method is rather inappropriate for the purpose of this research. First, focusing on revenues might be misleading by giving too much attention to the wrong regions since innovative ideas can arise independently from financial resources. Second, due to intra-regional differences (Shenkar, 2001), regional concentration doesn’t seem to be suitable for this study since every single country and nationality is important considering this study’s theoretical background.

22 Moderator variables

The relationship between organizational complexity and innovation performance is expected to be moderated by board characteristics and the ratio of emerging market subsidiaries.

One of the moderating factors is board size. It is measured by taking the total number of board members.

Another moderating factor is national board diversity. Because of its common use as a measurement instrument in team diversity studies, this thesis calculates national board diversity with the help of Blau´s Index (Blau, 1977). The index is defined as follows: B = [1 −Σ(Pi)²], with Pi = percentage of category i (nationality) in the group. The range of the index is zero to (k-1)/k. A score of 0 will represent a perfectly homogeneous group, whereas a score of (k-1)/k represents a perfectly heterogenous group. Therefore, the higher the value of the index, the higher the degree of national diversity within the board. Data about board characteristics is extracted from BoardEx and annual reports.

A last moderating factor, only affecting structural complexity, is the ratio of emerging market subsidiaries. This ratio is determined for every single company by dividing its number of subsidiaries located in emerging markets3 by its total number of subsidiaries.

Control variables

In this thesis, I will control for firm- and industry-level variables that are widely used in the literature on innovation performance and relevant to this study.

First, this thesis controls for firm age, measured by the difference between the current year in the data and the date of foundation. For age is controlled because prior studies have shown its effect on innovation. Older firms are related to less innovation because of less flexibility and fixed structures that impede the transformation and deployment of new knowledge (Capon et al., 1990; Naldi & Davidsson, 2014).

An additional control is financial performance, measured by return on assets (= net income / total assets). As suggested by prior studies, better performing firms with more financial resources tend to be more innovative and develop more products (Damanpour & Aravind, 2006).

Third, I control for firm R&D expenditures since there is a positive relationship between innovation input like investment in R&D and the innovation output like patents (Ahuja &

23 Katila, 2001; Leiponen & Helfat, 2010). To measure this variable, I take the natural logarithm of the total amount invested in R&D.4

Although the patent application process within Europe is similar there might be differences between the countries due to location advantages that influence the innovation performance. Therefore, this study controls for country of origin by using a dummy variable.

Since pressures to innovate differ between industries (Levin et al., 1985) I control for that by including a dummy variable based on the main activity of a firm at the 2-digit level of the NACE-code classification.

Research method

In the following, the methodological approach of this thesis will be outlined. In order to examine the relation between the variables, a regression analysis is performed. The dependent variable of this study is innovation output measured by granted patents, a count variable that takes only non-negative integer values. Thus, the assumptions of homoskedasticity and normally distributed errors that must hold for linear regression models are violated (Ahuja & Katila, 2001). Although count data with higher means tend to be normally distributed, wherefore an OLS regression would be a usable option, the data of this study showed violations of normal distribution and homoskedasticity even after re-modeling and adjusting the variables. Both, the Breusch-Pagan test and the White test for heteroskedasticity, show highly significant results, wherefore the 0-hypotheses stating constant variance and homoskedasticity need to be rejected (Appendix IV). Additionally, a Skewness-Kurtosis test, performed with the predicted residuals, leads to the rejection of the 0-hypothesis that the data is normally distributed (see Appendix IV).

A more appropriate approach for such data is a Poisson regression-type model (Henderson & Cockburn, 1996). Since the granting of a patent is an event and events usually occur rarely (Cameron & Trivedi, 2013), the distribution of the dependent variable might be heavily skewed towards zero. As seen in Appendix V, the skewness and kurtosis values, and a graphical illustration of the distribution of the dependent variable show that the data is not symmetrically clustered, with a small group accounting for most of the patents in 2016.

Although a log-linear Poisson regression seems to be an appropriate approach for analyzing count data (Hilbe, 2014), the underlying data in this study violates the very restrictive property of equidispersion. For Poisson regressions, the variance of the dependent variable must equal

24 the mean (Cameron & Trivedi, 2013). In this study, this is clearly not the case with a mean of 145.0458 and a variance of 276984.9 (Appendix V).

An alternative approach that can be used if the assumption of equidispersion is violated is a negative binomial model (NB). An NB has a less restrictive property and allows for greater flexibility in the variability of the variance errors (Hilbe, 2011). Running an NB regression includes a test for overdispersion. If alpha is significantly different than zero, the 0-Hypothesis that the data is Poisson distributed can be rejected. For the model of this study, this is the case (p=0.000). Therefore, an NB model is appropriate.

With zero events accounting for 37,4 % of the observations, we might have an excess zero problem with more zero data than the model predicts. In this case, a zero-inflated model would be the better choice (Hilbe, 2014) since it takes into account that not all zeros are the result of the same process. The zero-inflated NB is a mixed model approach where additionally to the NB model a logit model is performed to identify the origin of the zero events (Long & Freese, 2006). To check whether a zero-inflated model is preferred, a Vuong test was included which provides a test of the zero-inflated model against the NB model. Performing the test for the different models, the results have always been insignificant, therefore a zero-inflated NB is not the right choice and a negative binomial model is chosen.

In order to address the issue of unobserved heterogeneity, this study includes controls that capture firm-specific effects (Hilbe, 2014). However, there is still the possibility that the sample firms differ on unmeasured characteristics. If such unobserved heterogeneity is present and not controlled for, estimation problems can occur (Ahuja & Katila, 2001). As recommended by Cameron and Trivedi (2009), this study uses robust standard errors to attempt for heterogeneity in the model.

Another assumption that must hold for negative binomial models is that the observations are independent so that no autocorrelation is present. The cross-sectional nature of this study prevents such problem that might occur in longitudinal studies in form of within-firm correlation (Cameron & Trivedi, 2010).

25

Robustness checks

To find any misspecifications and to enhance the validity of the conclusions drawn from the NB regression, an original model can be additionally modified to check for robustness, in terms of removal or addition of regressors or the investigation of sub-samples (Lu & White, 2014). In this thesis, several robustness checks are conducted.

Firstly, the analysis is performed with a different set of control variables. Instead of controlling for ROA I will control for return on equity (ROE) to check whether the model is robust to the use of different financial indicators. Additionally, I will exclude the country dummy in a robustness check. Arguing that the patent application processes are similar between the chosen countries, excluding country as a control should not influence the findings.

Secondly, the robustness of the model will be tested by using a different independent variable that is expected to present the same indicator. Changing the measurement for firm size from a personnel indicator to a financial one is not appropriate as explained above. Thus, I change the measurement for structural complexity to geographical dispersion instead of counting the numbers of countries an MNE is present in. In order to capture the diversification of the network, the share of international subsidiaries in the whole subsidiary network is taken into account. Higher diversification means higher structural complexity. Since both variations capture structural complexity in terms of the subsidiary network, the model should be robust to the change of the measurement.

Thirdly, in order to check the sensitivity of the model regarding a different time lag, the time lag is changed. For this robustness check, I use a shorter time lag of just three years in line with prior studies on innovation performance (e.g. Alegre et al., 2006). I expect the firm characteristics of interest in this study hardly to change over a short period, wherefore the model should be robust to a different time lag.

A fourth robustness check uses standardized predictors in order to refute concerns of multicollinearity (Aiken, West, & Reno, 1991). Expected are changes in coefficient sizes, while signs and significance should remain the same.

26

RESULTS

Descriptive statistics

This section will introduce the data sample with the descriptive statistics. Table 1 below presents the overview of observations from 130 companies originating from five different European countries, operating in 30 different patent-intensive industries.

The discrepancy between the maximum (4313) and the minimum (0) value show that there are companies with a high patent output in the sample as well as companies with no patent output at all in 2016. Most of the companies can be found in the lower end of the spectrum as indicated by the relative positions of the mean and median values. A median of 2 shows that a minimum of 50% of the observed companies had a very low patent output in 2016.

Both independent variables, Firm Size and Subsidiaries have been transformed by natural logarithm to reduce the problem of NB models with extensive predictor variables (Feng et al., 2014). Regarding Firm Size, the minimum of 3.738 reflects 42 employed people. The largest company in our sample has 405,000 employees (log value 12.912). As indicated by mean and median, the majority of the companies is smaller than the average of the sample (20,576). Subsidiaries ranges from 0 to 4.625, with the minimum of subsidiaries in one foreign country while the maximum is 102. For this variable, the mean is nearly the same to the median resulting in an average of subsidiaries in 13 different countries. The independent variable used for robustness check Foreign_Ratio indicates that there are MNEs with all subsidiaries abroad. While the average has ca. 75% of its subsidiaries abroad, the minimum value is 12.5%.

With a minimum of 3 and a maximum of 29 members, there is a broad range of different board structures present. Regarding mean and median an average of 10.446 slightly exceeds the median of 9 members wherefore the majority of boards is smaller than the average, following the picture of the independent variables. Blau_Nat ranges from 0 to 0.777, indicating that there are completely homogenous boards as well as very diverse boards. Mean and median, however, show that most of the companies have less diverse boards in terms of nationalities. Regarding EM_Ratio nearly equal mean and median of 0.23 show that sample companies have more subsidiaries in developed than in emerging markets.

27 as country, are dummy variables with the values showing that most of the companies in the sample are based in manufacturing industries and originating from Germany.

Table 1: Means, standard deviations, median, minimum and maximum values

Correlation

The correlation matrix for the variables used in this study can be seen in table 2. Values lower than 0.1 represent no effect, from 0.1 to 0.3 a small, from 0.3 to 0.5 a medium and above 0.5 a high effect. Values above 0.8 are an indicator for multicollinearity and correlations exceeding 0.9 represent a harmful level of correlation (Tabachnick & Fidell, 1996). The basis for the evaluation is the test for variance inflating factors (VIF). Acceptable values in order to be able to enter the regression simultaneously should be below 10 for VIF (Myers, 1990; Field, 2009). Since none of the variables exceeds this value, all variables can be used for the regression analysis (see Appendix VI).

A high correlation of 0.839 exists between ROA and ROE. Since those variables do not enter the same model, this correlation is not problematic. Another very high correlation can be found between Firm Size and Subsidiaries. Since both variables try to capture the concept of organizational complexity a high correlation between those two variables was expected beforehand. However, there were thoughts about dropping one of the variables since entered separately in the regression model, both have a significant effect, while entered together only Firm Size has a positive and significant impact (see Table 3). However, the log likelihood for the model including both variables is better wherefore I reject to drop one of the variables.

28 larger and more complex firms seem to have larger boards. However, the height of this correlation is not problematic.

More problematic could be the high correlation of R&D Expenditures with three other variables. Since previous research showed that patents and R&D expenditure are substitutes (Hagedoorn & Cloodt, 2003) and including R&D expenditures probably take away explanatory power, I decided to drop this variable. The mean VIF also improves when dropping R&D Expenditure (Appendix VI).

An interpretation of the coefficients regarding the effects of the variables on patent output is however not appropriate since they only present correlation but are not valid predictors. The interpretable regression results are presented in the following section.

Table 2: Correlation Matrix

Regression results

The results of the NB regression are presented in Table 3. To test the seven hypotheses this thesis uses eight different models, introducing the variables successively. While models (1) - (6) demonstrate only direct effects of the variables, models (7) and (8) include interaction terms and are therefore the primary models used for interpreting the results.

Model (1) only includes the dependent variable and controls. The influence of the industry and country dummy is significant and negative, meaning that companies from Germany operating in the manufacturing sector have significantly higher innovation performance than other companies indicating that there are indeed different pressures between industries and locational advantages. The coefficient for age is positive and significant on a 0.01% level, which contrasts with findings of previous studies (e.g. Naldi & Davidson, 2014). ROA is the only control with insignificant results.

29 Models (2) and (3) introduce both independent variables separately. In model (2), the highly significant (p<0.01) and positive impact of firm size on innovation performance is outlined. As expected, firm size positively affects innovation performance, therefore supporting hypothesis 1. In line with the predictions, model (3) shows a positive and significant relationship between structural complexity and innovation performance, thus supporting hypothesis 2. However, as the models are incomplete, the results of those two models must be interpreted with caution.

Model (4) introduces the moderating variables and tests their direct impact on innovation performance. While the impact of the emerging market ratio is, contrary to the predictions, negative but not significant, board size and national diversity both have a positive and highly significant (p<0.01) effect. A larger and more diverse board thus seems to positively affect innovation performance in European MNEs.

In model (5) both independent variables are introduced together, leading to different results than models (2) and (3). While the coefficient for company size increases and remains highly significant, the coefficient for structural complexity decreases turns negative and loses its significance. Those results, maybe caused by the high correlation between both variables, lead to a further support for hypothesis 1. Hypothesis 2 however, cannot be supported any longer.

Model (6) presents the direct effects of both independent and moderating variables. The results of the sixth model corroborate the results of models (4) and (5) with significance levels largely remain the same, except for board diversity where the significance slightly decreases from 0.01% level to 0.05% level. Additionally, the coefficients of all variables slightly decrease.

Finally, regarding models (7) and (8), the results of the leading regression line with the highest predictive power, including all variables and interaction terms are presented. A log likelihood ratio closest to 0 for model (7) supports the notion that the complete model has the highest goodness of fit. Outcomes of the previous models regarding firm size are confirmed revealing a significant and positive relationship with innovation performance. Therefore

Hypothesis 1 is supported. However, including interaction terms diminished the significance

of firm size, especially the interaction with board size in model (7).

Including the interaction terms changed nothing regarding the stand-alone effect of structural complexity. Its direct effect remains insignificant. Models (7) and (8) find no evidence in favor of the proposed relationship between structural complexity and innovation performance. Thus, Hypothesis 2 is not supported meaning that more locations do not necessarily lead to more innovation.