International diversification and the
impact of R&D on corporate
performance
A comparison made in several developed countries
with different economic systems
Master Thesis
of
Camillo Valentin Werdich
At the Department of Economics and Business of RUG
Academic supervisors:
Dr. H.J. (Rian) Drogendijk
-Information
Camillo Valentin Werdich
This master thesis reexamines the relation between international diversification and R&D intensity on corporate performance. Further we point out that the home coun-try environment influences the innovative behaviour of firms. The importance of international markets and the role of innovation has increased dramatically over the last two decades especially for Multinational Companies from developed countries and academic research has increased concurrently. In this context, home country environments represent an important issue for innovation. Our findings provide sig-nificant evidence for an inverse U-shaped relationship between R&D intensity and corporate performance. In the same context an inverse U-shaped relation was also found for international diversification, however, without significance. The interplay between the non-linear R&D intensity and international diversification effects prog-nosticate significant positive influence on corporate performance. By applying a multilevel approach we found new results regarding the role of home country envi-ronments and economic systems in terms of innovation. However, the latter has only a marginal effect on innovative behaviour of firms. Finally, we provide new insights into the relation between radical and incremental innovation and macroeconomic issues.
Key words:
Contents
Information iii
Abstract iv
List of Figures vii
List of Tables viii
List of Abbreviations ix
1. INTRODUCTION 1
2. THEORY AND HYPOTESES 5
2.1. Research and Development intensity and Corporate Performance . . . 5
2.2. International Diversification and Corporate Performance . . . 7
2.3. Impact of International Diversification and R&D intensity . . . 10
2.4. Impact of Different Economic Systems . . . 11
2.5. Institutional Environment and Industry-specific Advantages . . . 13
2.6. Conceptual Model . . . 19
3. EMPIRICAL ANALYSIS 20 3.1. Data and Sample Design . . . 20
3.2. Dependent Variable . . . 22
3.3. Research and Development Intensity . . . 22
3.4. International Diversification . . . 22 3.5. Economic system . . . 23 3.6. Control Variables . . . 24 3.7. Statistical model . . . 24 3.8. Modelling Procedure . . . 25 3.9. Estimation Method . . . 26
3.10. Evaluation of Method Assumptions . . . 27
4.6. Results of Multilevel approach . . . 35
4.7. Robustness check . . . 39
5. CONCLUSION & DISCUSSION 41 5.1. Managerial implications & Added Value . . . 43
6. LIMITATIONS & FUTURE RESEARCH 45 Bibliography 47 APPENDIX 54 A. 5 spheres for determination of LME and CME countries . . . 54
B. Differences between LME and CME . . . 55
C. Total Early-Stage Entrepreneurial Activity . . . 56
D. Modelling Procedure . . . 57
E. Estimation results . . . 58
List of Figures
1.1. Illustration Multilevel . . . 3
2.1. Relation Research and Development . . . 7
2.2. Relation International Diversification . . . 8
2.3. Hofstede dimensions . . . 16
2.4. Early-Stage Entrepreneurial Activity rates (TEA) within age groups in 2014, by geographic regions . . . 17
2.5. Entrepreneurial Employee Activity . . . 18
2.6. Conceptual Model . . . 19
4.1. Moderation Effect of R&D intensity on the relation between ID and ROA . . . 32
4.2. Variance components model . . . 35
4.3. Random Intercepts . . . 37
4.4. Random Slope . . . 38
B.1. Differences between LME and CME . . . 55
C.2. Total Early-Stage Entrepreneurial Activity . . . 56
3.1. Sample criteria . . . 20
4.1. Multicollinearity . . . 30
4.2. Normality . . . 30
4.3. Descriptive Statistics - Correlations, Means and Standard deviations . 33 4.4. Results of Regression Analysis . . . 34
4.5. Results of HLM . . . 39
A.1. 5 spheres . . . 54
D.2. Modelling Procedure . . . 57
F.3. Results of Robustness check - reduced regressors . . . 60
F.4. Results of Robustness check - Sample . . . 61
List of Abbreviations
BLUE Best linear unbiased estimation CMEs Coordinated market economies EEA Entrepreneurial Employee Activity FDI Foreign Direct Investment
FSA Firm specific advantage
GEM Global Entrepreneurship Monitor HLM Hierarchical Linear Modeling IT Information technology LMEs Liberal market economies MNEs Multinational Enterprises OLS Ordinary least squares PPP Purchasing Power Parity R&D Research and Development RBV Resource-based view
SBIR Small Business Innovation Research STTR Small Business Technology Transfer TCE Transaction cost economics
TEA Total Early-Stage Entrepreneurial Activity VIF Variance inflation factor
1. INTRODUCTION
Research and Development (R&D) and international diversification represent essen-tial factors for firms’ strategy (Hitt, Hoskisson and Kim, 1997; Wan and Hoskisson, 2003; Alessandri and Seth, 2014). In the recent years, the role of R&D, multinational activities and the impact of institutional environment on firms emerged in empirical literature (Hitt et al., 2006; Kirca et al., 2011; Gentry and Shen, 2013). Scholars, i.e. Hitt et al. (2006); Wan and Hoskisson (2003) and Lu and Beamish (2004) empha-sised the interaction between R&D intensity and international diversification and pointed out future profits in firms performance. However, managers are facing many challenges in connection with R&D investments and the right level of international diversification.
For instance competitors from emerging markets offer comparable products at lower prices and western companies struggle to keep pace. One of the key drivers in this context is innovation. Prior research shows that global challengers increasingly see the need to become rapidly more innovative. For example 46 % of Huawei’s employees are in R&D, while Mindary generates more U.S patents per revenue dol-lar than western competitors do (Shaughnessy, 2013). Moreover, according to the U.S. Patent and Trademark Office, the patents growth rate increased more than three times faster in rapid developing countries than in others (Shaughnessy, 2013). However companies from developed countries are still holding a strong position re-garding innovations. Experts argue that, ”Most technology ventures in China are of the C2C variety, i.e., Copy-To-China. It’ll take at least a couple of decades before we see an Apple or a Google emerge here,” (Gupta and Wang, 2011, p.1). Therefore, it would be interesting to analyse the role of the home country regarding innovative behaviour.
and in an international context, R&D investments can be even more profitable for Multinational enterprises (MNEs) (Hitt, Hoskisson and Kim, 1997; Hagedoorn, 2002; Parcharidis and Varsakelis, 2010). International diversification can push the yielding R&D intensity and therefore increase firm performance even more (Kotabe, Srini-vasan and Aulakh, 2002; Lu and Beamish, 2004). This phenomenon is mostly related to the internalisation theory which has become a leading perspective in both interna-tional business and in management strategy literature (Kirca et al., 2011). Therefore, MNEs would be well advised to engage in R&D and international diversification and to find the most profitable level of both (Lu and Beamish, 2004).
However, the world is not flat; the strategic behaviour of firms is influenced by different externalities. These are not only individuals but also producer groups and governments (Hall and Soskice, 2001). Thus, the home country environment has an important influence on firms’ strategies as well as their organizational structure (Wan and Hoskisson, 2003). Hall and Soskice’s ”Varieties of Capitalism” (VoC) approach defines the differences among developed countries (Hall and Soskice, 2001). The main distinction consists of two economic systems, i.e. liberal market economies (LME) (e.g., U.S., Canada, U.K., Australia) and coordinated market economies (CME) (e.g. Germany, Sweden, Japan, Austria). One of the key findings in this study are the country specific advantages regarding innovation. The VoC approach differentiates between radical and incremental innovation. Furthermore, Hall and Soskice emphasise sector specific advantages within LME and CME for certain types of innovation. The specific characteristics and differences will be explained in more detail in section 2.4.
3
the generalizability is limited (Hitt et al., 2006). Another research gap is the lack of clarity regarding the VoC approach, because Hall and Soskice’s findings are based on relatively vague assumptions (Taylor, 2004; Akkermans, Castaldi and Los, 2009). The critical factors will be described later in more detail.
Apart from that we found two major problems in the existing literature concern-ing VoC. First, the VoC approach as well as the critical stream of literature used patents as a measurement for innovation (i.e. Hall and Soskice (2001); Taylor (2004)). Thereon they draw conclusion on specific economies. Both from a the-oretical point of view and on grounds of new findings regarding product piracy we consider the measurement of patents as not appropriate in this context. Patent registration behaviour of firms has changed over the last decade because of counter-feiting. Especially knowledge based firms, like the machinery manufacturing sector tend to register fewer patents than in the past (Spiegel, 2008). Due to the disclosure requirements of 18 months (German law) after registration, competitors can easily avail themselves of the new technology. Thereby firms, in places, give up their com-petitive advantage (Spiegel, 2008). This is why the patent-based measurement for innovation is questionable. Consequently, we use R&D intensity as measurement for innovation. Also the existing findings are based only on firm level data (i.e number of patents). Furthermore only single level analyses are applied, which leads to the assumption that the predictor variables are independent. However this is not the case for R&D investments. This in turn leads to an ecological fallacy.
In contrast, let us suppose that home country institutions matter for R&D and has an influence on the strategic behaviour of firms and are not independent as assumed in prior studies. Therefore we will apply a multilevel analysis which considers more levels, (1) firm level and (2) country level formal (regulations) as well as informal (culture) institutions which are categorised as (3) LMEs and CMEs as illustrated in the following figure 1.1.
The aim of this master thesis is to determine the performance effects of different degrees of international diversification and R&D intensity as well as the effect of the macro institutional environment on firms’ innovational behaviour. It raises the following research question:
(1) To what extent can firms leverage R&D investments in order to increase their profits from international operations and how does the interplay between interna-tional diversification and R&D investments affect firm performance? (2) Is the home country a relevant issue for innovation and is there a difference between economic systems?
In order to answer the research question we refer to different theories from sev-eral disciplines (i.e. international management, economics, international business and strategic management). More specific theoretical domains which have been used are Internationalization theory, Resource Based View (RBV), Transaction Cost Economies (TCE), Varieties of capitalism (VoC).
The results of the thesis contribute to strategic decision making process of MNCs and to the academic audience. The results are especially relevant for the subject area of ”Multinational Corporations and Internationalization” with the focus on firm strategy. However, the theoretical findings are also relevant for other disciplines like marketing, finance, entrepreneurship and economics (Hitt et al., 2006). Our con-tribution submits three theoretical perspectives. First, this paper provides new findings as to the effect of R&D intensity on the non-linear relation between inter-national diversification and firm performance. Second, the inverse U-shaped impact of international diversification will be further elaborated on the bases of the new findings, thus contributing to the generalisability with results from non-U.S. com-panies and a more diverse sample than most of the prior studies. Third, this thesis will examine the impact of institutional environment on R&D intensity by apply-ing a multilevel analysis. At the same time we test the VoC assumptions. For the managerial perspective, our results will provide an approach for understanding the relation between international diversification and R&D activities and for assessing the impact of their home environment. Thus, the decision making process regarding internationalization and determination of R&D investments will be facilitated for the management. Additionally, this study can be used as a starting point for further in-vestigations respectively, to consider the influence of institutional environment from a new econometric perspective (Multilevel).
2. THEORY AND HYPOTESES
Before we start discussing the impacts and interaction of the variables it is necessary to develop a fundamental understanding of the single factors. Therefore we review the existing literature for each variable and conclude each section with our expec-tations and the hypotheses which will be tested later on. We start with reviewing R&D and discuss why R&D is an essential key factor for the long term success of a firm and link it to innovation. Next, we will review and determine the impacts of in-ternational diversification and differentiating the latter from inin-ternational activities based on Hitt, Hoskisson and Kim (1997). Thereon, we will discuss the interaction of these issues. This is followed by an introduction to the varieties of capitalism ap-proach and the macro effects, which in turn are expected to influence the micro level empirical results. Finally, we will present a conceptual model which summarizes our expectations.
2.1. Research and Development intensity and Corporate Performance R&D activities are crucial to firm performance and have been intensively studied over the last decades (Parcharidis and Varsakelis, 2010; Bracker and Ramaya, 2011; Song et al., 2015). The impact of R&D has been studied at several different levels. According to Wieser (2005) the prior research focused mainly at macroeconomic, industrial, sector and firm level.
R&D is frequently used as measurement for different factors. For instance Hitt, Hoskisson and Kim (1997) used R&D intensity as a proxy for innovation. Kirca et al. (2011) used R&D as a variable for firm specific assets and Wieser (2005) as measurement for productivity growth. The basic idea is that R&D adds value to the firm, generating a competitive advantage which in turn is likely to increase the financial performance of a firm.
this study focuses only on manufacturing and IT firms because the results are likely more decisive than those obtained form the low-technology or service industry. This coincides with the results of the meta analysis of Kirca et al. (2011).
Consistent with Wieser (2005), we found large and significant impact of R&D in-tensity1 on firm performance in the existing literature (Bardhan, Krishnan and Lin,
2013; Kotabe, Srinivasan and Aulakh, 2002; Bracker and Ramaya, 2011). However, the estimated returns on firm performance vary between different studies.
According to Hagedoorn (2002), R&D is devoted to increase the firms’ scientific and/or technical knowledge as well as the application of that knowledge in order to create new improvements in products or processes. In essence, R&D improves corporate performance through two channels. First, improvement of the product design and development. Second, improvement regarding the manufacturing process, in terms of efficiency (Parcharidis and Varsakelis, 2010). The resulting consequences from both channels generate theoretically high innovative outputs. At the same time firms are able to take advantage of economies of scale (Artz et al., 2010).
According to Hitt, Hoskisson and Kim (1997), the relationship between R&D in-tensity and firm performance is positive. The proposed relation is caused by the advantages mentioned above. However more recent studies (i.e.Bracker and Ramaya (2011)) show significant evidence for a curvilinear relationship rather than a lin-ear relation between R&D and firm performance. The existing literature provides different explanations for this phenomenon.
Artz et al. (2010) argues that a possible economic inefficiency can arise due to firm size and/or firm age. Firm size has a strong influence on decision making, resource allocation or coordination process. Consequently, we can assume that large firms need more time for invention than smaller firms (Adams and Brock, 1986). Other researchers argue that with increasing firm age conservatism of the management becomes more likely (Artz et al., 2010). In other words, the group of founding entrepreneurs were often replaced by professional managers which in turn lowers the effect of R&D on firm performance because other factors become more important. Thus, the assumption is that at a certain point the costs of R&D exceed the returns, which of course has a negative influence on firm performance too. The resulting relation is thereby expected to be inverse U-shaped.
Consistent with this empirical evidence of Bracker and Ramaya (2011), the following hypothesis posits an inverse U-shaped relationship between R&D intensity and firm performance.2
1R&D intensity is defined as R&D / Operating Turnover
2.2. International Diversification and Corporate Performance 7
Hypothesis 1: There is a inverse U-shape relationship between R&D intensity and firm performance.
The following figure 2.1 illustrates our expectations.
Figure 2.1.: Relation Research and Development, Source: Own model
2.2. International Diversification and Corporate Performance
In the existing literature international diversification is often labelled to or asso-ciated with geographic diversification, international expansion or multinationality (Lu and Beamish, 2004; Wiersema and Bowen, 2011; Alessandri and Seth, 2014). All these designations tend to refer to the same strategic management construct, namely international diversification (Hitt et al., 2006). International diversification is defined as the ”expansion across the borders of global regions and countries into different geographical locations and markets” (Hitt, Hoskisson and Kim, 1997, p. 767). In international business as well as in strategic management literature it be-came a leading perspective (Kirca et al., 2011). According to the internalisation theory, Multinational Enterprises (MNEs) can increase their profits by transferring their intangible assets across national borders (Buckley, 1988; Kirca et al., 2011). According to Hitt et al. (2006) international diversification provides several advan-tages which are similar to product diversification and other strategies. Advanadvan-tages like economies of scale and scope, access to new resources, knowledge acquisition, location advantages and extension of innovative capabilities as well as risk spreading regarding revenue fluctuation and investment risks are often associated with inter-national diversification (Hitt, Hoskisson and Kim, 1997; Kim, Hwang and Burgers, 1993; Tallman and Li, 1996; Bathelt, Malmberg and Maskell, 2004). These ad-vantages can be derived from four motivations, which are essential for firms to ex-pand abroad or consider foreign direct investment (FDI) (Dunning, 1998): Resource-seeking, Market-Resource-seeking, Efficiency-Resource-seeking, Strategic assets-seeking.
investments the management faces additional challenges related to a new operation, for instance, staffing, installing facilities or establishing internal management systems (Lu and Beamish, 2004). These liabilities lower the advantage of more international operations, especially for a more international diverse enterprise.
The previous research has shown very inconsistent results regarding the effect of international diversification on firm performance (Lu and Beamish, 2004; Wiersema and Bowen, 2011). According to Tallman and Li (1996) previous research supports positive as well as negative linear effects. More recent studies propose curvilinear effects (Hitt, Hoskisson and Kim, 1997; Kirca et al., 2011). However, the results show a wide variation of curvilinear models between international diversification and firm performance, inverted U-shaped as well as horizontal S-shaped curve relationship (Lu and Beamish, 2004).
The horizontal S-shaped and the inverted U-shaped relation have the same funda-mental basis, which will be explained later in more detail. The horizontal S-shape assumption additionally shows that low levels of international diversification show the same effects as too high levels, demonstrating its negative effect on firm perfor-mance (Hitt et al., 2006). However, at a very low level of international diversification, the negative effect can be explained by lack of experience. The following figure 2.2 illustrates on the one hand the horizontal S-shape relationship and on the other hand it indicates the inverse U-shape effect which we expect in our study.
Figure 2.2.: Relation International Diversification, Source: Own model
2.2. International Diversification and Corporate Performance 9
in the field of international business and management.” (Wiersema and Bowen, 2011, p.152)
The more in-depth analysis of the existing international business and strategy lit-erature provides an explanation of international diversification as a result of two distinct bodies of literature, which refer to the RBV and the TCE (Lampel and Giachetti, 2013).
The benefits of international diversification are mostly based on the RBV theory. The RBV assumes that it is more profitable for firms to diversify their operations by transforming internal capabilities and core competencies to other markets (Hitt, Hoskisson and Kim, 1997). For instance the meta analysis of Kirca et al. (2011) indicates that multinationality provides an efficient organizational form that enables firms to transfer their firm specific advantage (FSA) across borders which leads to higher returns and in turn increases firm performance (Hitt et al., 2006). Moreover prior research in the field of strategic management found that the scale and scope of a firm’s international diversification indicate three important advantages (Hitt et al., 2006): (1) Market power of a firm, (2) effectiveness of resource usage and (3) access to abundant resources.
The second body of the literature refers to TCE. According to TCE theory, compa-nies’ organizational costs increase rapidly the higher the international diversification (Lampel and Giachetti, 2013). This for example, can be caused by the different mix of threats and opportunities which varies across different countries, as men-tioned above. Unexpected fluctuations in exchange rates, trade barriers, cultural differences, logistical costs or unstable institutional environment as well as agency problems can rapidly lead to higher costs (Hitt et al., 2006). These factors are likely to increase transaction costs considerably, likewise the managerial information-processing demands. Thus, we can say that international diversification is complex to manage and can also impair the above mentioned benefits (Hitt, Hoskisson and Kim, 1997; Lampel and Giachetti, 2013).
An examination of these two theories suggests that the proposed benefits become outperformed at a certain level of international diversification. In other words, the arguments of both theories depend on the level of international diversification. This implies that especially in the downside related section (high level of international diversification) TCE arguments become more important. This in turn results in a inverse U-shaped relationship.
However, more recent studies are using the Herfindhal index as a measure of inter-national diversification because such an index also takes into account the number of geographic segments and the proportion of value that is derived from each region. Thus we emphasise here the importance of differentiating international activities and international diversification not only by definition but also by measurement.
Based on the existing literature and the described RBV and TCE evidence, we expect an inverse U-shaped relationship between international diversification and firm performance. In other words, we expect a positive trend effect up to a certain point. Beyond, the associated costs outperform the potential benefits (Lampel and Giachetti, 2013). A horizontal S-shaped relation is not relevant in our context, because we analyse only large and very large companies. Thereby, we can assume that these companies already provide a certain level of international diversification which exceeds the proposed low level effects of Lu and Beamish (2004). This leads us to the following hypothesis.
Hypothesis 2: There is an inverse U-shaped relationship between the level of inter-national diversification and the level of firm performance.
Other international business scholars have shown that geographic expansion can trig-ger different inferences about the net benefits of internationalization (Hitt, Hoskisson and Kim, 1997; Lu and Beamish, 2004). This leads us to the importance of R&D and the interaction effects in this context.
2.3. Impact of International Diversification and R&D intensity
2.4. Impact of Different Economic Systems 11
Hypothesis 3: Research and Development intensity moderates the relationship be-tween international diversification and firm performance in such a way that high lev-els of research and development intensity increase the performance gains attributable to international diversification.
2.4. Impact of Different Economic Systems
Institutional environments are one major stream in the field of international business research. Scholars agreed that different institutions matter for corporate strategy and performance. However, the very meaning of institutions is still controversial and remains defined by a set of approaches only (Jackson and Deeg, 2008). The existing literature provides a wide range of both theoretical and methodological approaches to studying institutions. Furthermore, the scholars draw different conclusions from diverse fields of science such as: Social science like economics (North, 1990; Mayer and Whittington, 2003), political science (Immergut, 1998; MacKinnon et al., 2009) and sociology (Powell and DiMaggio, 2012; Streeck and Thelen, 2005). Despite the interdisciplinary cross-fertilization the impact on firm performance is still unclear. We follow North (1990) regarding the definition of institutions. Institutions are the ”rules of the game” which are either formal or informal. Formal institutions refer to organizations of the government and public services. Informal institutions are related to informal rules, peer pressure and personal networks. Both formal and informal institutions are setting boundaries for the action of firms. However, they reduce uncertainty and foster the predictability of unfamiliar actors’ behaviour to facilitate exchange at the same time. In order to get a fundamental knowledge how institutions affect organizational operations, Scott (1995) provides three pillars. (1) Regulatory, which describes the formal institutions. For instance the existing laws and rules of a particular nation. (2) Cognitive, which can be seen as the informal component. It represents the cognitive structure and social knowledge of the people in a given country. It describes how people approach, select and interpret information. (3) Normative, which is also related to informal institutions, describes social norms, beliefs, values and assumptions about human nature and human behaviour. Important results in the context of normative aspects are the findings of Hofstede’s ”Cultural ComPass” (Hofstede, 2015).
Hall and Soskice (2001) elaborated this framework to facilitate understanding sim-ilarities and differences among developed countries. The basic element of this ap-proach is that the political economy is actor-centred. This implies that the political economy is a field of multiple actors; each player seeks to advance his own interests in a rational way. This results in a strategic interaction with others. The most relevant actors in this context are firms, individuals, producer groups and governments. In pursuing their goals they act as key agents of change in the face of international com-petition or technological change, the result of which is the overall level of economic performance (Hall and Soskice, 2001). Thus, we see firms in this context as actors seeking development and exploitation of their core competencies. Furthermore, they use their dynamic capabilities as capacities for producing and distributing goods or services profitably.
Thereon, Hall and Soskice (2001) divide national political economies into liberal market economies (LMEs) (e.g., U.S., Canada, U.K., Australia) and coordinated market economies (CMEs) (e.g. Germany, Netherlands, Sweden, Japan, Austria). A frequently mentioned example is the difference between the U.S. and Germany. Both have highly developed economies, however, a more in-depth analysis shows that these economies differ in several economic institutional factors. For instance, level of social protection, availability of collective goods (social solidarity of a nation) and the distribution of income (Hall and Soskice, 2001).
The distinction between LMEs and CMEs is based on five spheres in which MNEs must develop relationships in order to solve coordination problems central to their core competencies. The five spheres are defined in detail in Appendix A.
2.5. Institutional Environment and Industry-specific Advantages 13
Based on these results, we continue with the same clusters as Hall and Soskice (2001) did regarding economic systems and the proposed sector specific advantages, which is also in line with Akkermans, Castaldi and Los (2009).
2.5. Institutional Environment and Industry-specific Advantages
According to Hall and Soskice (2001), each type of capitalism has its own industry-specific technological advantage as already mentioned. The key distinction, which is especially important in the context of this study, is between radical and incremental innovation. Radical innovation ”entails substantial shifts in product lines, the devel-opment of entirely new goods, or major changes to the production processes” (Hall and Soskice, 2001, p. 38-39). Incremental innovation is ”marked by continuous but small-scale improvements to existing product lines and production processes” (Hall and Soskice, 2001, p. 39). Moreover, Hall and Soskice hypothesized that LMEs are specialized in radical innovation and that CMEs are specialized more in incremental innovation.
Radical innovation is particularly important for fast-moving technology industries such as biotechnology and software development industries which require rapidly changing innovative design and product development. According to Akkermans, Castaldi and Los (2009), especially in complex system based industries like telecom-munications, defence systems as well as entertainment radical innovation is decisive. Incremental innovation is characterised by keeping focus on maintaining competi-tiveness in the production of goods. This is essential for industries like consumer durables, transport equipment or machine tools. The key factor for this kind of industry is to maintain the high quality of an established product and secure con-tinuous improvements in their production process. So innovation takes place incre-mentally and not radically (Akkermans, Castaldi and Los, 2009; Hall and Soskice, 2001).
However, more recent studies provide very controversial arguments against the VoC approach (Taylor, 2004). According to Schneider and Paunescu (2012), the influence of the VoC approach is surprisingly strong, because the empirical validity of the research is still unclear.
Akkermans, Castaldi and Los (2009) maintain that the hypothesis3 regarding innova-tion should be rejected. Akkermans, Castaldi and Los (2009) state that the empirical analysis tells more about economic specialisation patterns rather than about techno-logical specialisation. In our option this assumption is reasonable because, Hall and
Soskice studied only two periods i.e. 1983-1984 and 1993-1994 and the patent spe-cialization by technology classes, and this only in two countries, U.S. an Germany. Hall and Soskice found that the European Patent Office provided relatively more patents in dynamic technologies than the US inventors. Furthermore, the results show that patents are more related to incremental innovations in Germany than patents in the US.
However, Akkermans, Castaldi and Los (2009) and Taylor (2004) still found a sig-nificant difference in their results. Taylor (2004) concludes that Hall and Soskice’s predictions do not stand up to the empirical model used. The main difference here was the inclusion of patents weighted by forward citation, and scholarly publica-tions compared to VoC approach which just consider the simple amount of patents. Akkermans, Castaldi and Los (2009) in contrast found that LMEs are slightly more specialized in radical innovation in industries like electronics. CMEs showed the same effect for machinery industries. To summarize the existing findings, we agree with Akkermans, Castaldi and Los (2009) that the truth is somewhere in between of Hall and Soskice’s and Taylor’s findings, as presented above.
Thus, we can say that industry-specific advantages regarding innovation in differ-ent institutional environmdiffer-ents are still unclear. Thereby the question arises which industry sectors gain more from LMEs or CMEs environment?
We follow Akkermans, Castaldi and Los (2009) regarding our expectations. We expect that LME are more supporting for the IT industry, while CME are more profitable for the manufacturing industry.
To answer this question we first have to fully understand the possible differences between the economies and the fundamental theoretical concepts of the previous studies. From a theoretical point of view, both the VoC approach and the criticised body of the literature consider only data on firm level. For example they used patents of firms in a specific-industry, patents weighted by forward citations or scholarly publications to determine country specific advantages (Taylor, 2004). Thereon, they justify their findings on a macro level, specifically on country- / industry specific advantages.
2.5. Institutional Environment and Industry-specific Advantages 15
key variable to determine the incentives for R&D activities and for the behaviour across nations. Wan and Hoskisson (2003) argue that the home country environment has an important influence on firms’ strategies and their organizational structure. Not only may it influence stakeholders’ satisfaction or the decision making process of a firm, it also affects the organizational innovation, according to Wan and Hoskisson (2003).
Due to the introduced ”Rules of the Game” by North, the informal and formal in-stitutions are partly measurable and explainable. Based on that, scholars compared these institutions along different countries. In the following we provide an overview covering the most important countries.
To start with the formal differences. A fundamental difference lies in the legal systems of Canada and Germany for instance. The Canadian legal system is based on common law, as in the UK and USA. In contrast Netherlands, Germany and most of the other EU countries as well as Japan and China are based on a civil legal system. The main difference between these systems is that in civil law systems codified statures predominate, whereas in common law countries, case law is of primary importance, in the form of published judicial opinions. According to the World Bank (2015a), the following key differences are characterised by five features which affect companies most: (1) Written constitution,(2) Judicial decisions, (3) Writings of legal scholars, (4) Freedom of contract and (5) Court system applicable to purchasing power parity (PPP) projects.
Furthermore there are also differences within legislation. According to Knoke (1996) all governments take increasingly influence on the growth and trade strategies of firms by implementing laws and regulations in order to facilitate economic growth, which are at the same time, intended to protect their economy.
new technologies. In conclusion, we can say that thereby firms from CMEs are less flexible than their competitors from LMEs and therefore radical innovation is more likely to happen in LMEs. Moreover governments support different initiatives to promote high-tech innovations. Formal government programs like the Small Busi-ness Innovation Research (SBIR) or Small BusiBusi-ness Technology Transfer (STTR) programs in the United States can foster innovation within a certain country. In the UK one important program is the Science Enterprise Challenge. The European Commission, too, implemented such initiatives, such as the ProTon Europe and the European Knowledge Transfer Association in order to support innovation (Autio et al., 2014). More specifically, the U.S. federal government introduced in 2011 a large supporting program for the clean technology sector (Hargadon and Kenney, 2012). Such formal initiatives can increase the variation across different nations and therefore trigger R&D intensity of firms.
Informal differences between LMEs and CMEs can further be partly derived from Hofstede (2015) dimensions. The results of Hofstede (2015) show that USA, U.K. and Canada are more similar regarding power distance, individualism, masculin-ity, uncertainty avoidance, long term orientation and indulgence than countries like Germany, Netherlands or Japan. The following figure 2.3 presents the informal dif-ferences for the chosen sample. The curve shows a significant difference between uncertainty avoidance and long-term orientation between LMEs (blue area) and CMEs (green area) (Hofstede, 2015).
Figure 2.3.: Hofstede dimensions, Source: (Hofstede, 2015)
2.5. Institutional Environment and Industry-specific Advantages 17
the efficiency of operations and hamper the decision making process regarding inter-national expansion (Jackson and Deeg, 2008). However, cultural differences can also play an important role regarding a nations’ innovation and entrepreneurial activity. The Global Entrepreneurship Monitor Report (GEM) presents entrepreneurial at-tributes and activities on both an individual and a global level. The report provides insights into the patterns and trends from two perspectives: (1) geographic regions and (2) economic development stages. In our context only the geographic regions are relevant and therefore, we will discuss these issues in greater detail as follows. The Total Early-Stage Entrepreneurial Activity rate (TEA) can be seen as an infor-mal institutional difference. TEA is defined as: ”Total Early-Stage Entrepreneurial Activity rate (percentage of individuals aged 18-64 in an economy who are in the pro-cess of starting a business or are already running a new business, not older than 42 months).” (Singer, Amor´os and Arreola, 2014, p.35) Furthermore the author argues that ”this indicator can be additionally enhanced by providing information related to inclusiveness (gender, age), impact (business growth, innovation, international-ization) and industry (sectors)”(Singer, Amor´os and Arreola, 2014, p.35)
The following figure 2.4 shows that US and Canada’s society has a higher TEA rate than their European peers. This in turn indicates that the individuals of LMEs (north America) are more engaged in innovation and in internationalization, this in turn increases the institutional variation. A detailed overview of the specific countries is shown in appendix C in figure C.2.
Figure 2.4.: Early-Stage Entrepreneurial Activity rates (TEA) within age groups in 2014 by geographic regions,Source: (Singer, Amor´os and Arreola, 2014, p. 43)
the employee engagement in innovation. The Entrepreneurial Employee Activity (EEA) rate is defined as the ”involvement of employees in entrepreneurial activities, such as developing or launching new goods or services, or setting up a new business unit, a new establishment or subsidiary.” (Singer, Amor´os and Arreola, 2014, p. 23) Again, here we can see that the US, Canada and UK scores are higher than those of Japan and Germany as presented in the following figure 2.5.
Figure 2.5.: Entrepreneurial Employee Activity, Source: (Singer, Amor´os and Arreola, 2014, p. 51)
Based on these findings we can assume that the employees of LMEs are more likely to engage in innovation than their peers form CMEs. Consequently we can expect that the R&D intensity in LMEs is higher than in CMEs. This in turn would again increase the institutional variation.
As outlined before, formal and informal institutions shape country specific charac-teristics which is the main source of the proposed variation. Based on that we can assume that the home country matters regarding R&D intensity and international diversification. There is only limited literature regarding our expectations, also men-tioned by Hitt et al. (2006). One notable contribution in this field of study comes from Wan and Hoskisson (2003). Their results show that munificent home country environments moderate the relationship between international diversification and firm performance. However, this study considers only Western European countries. Therefore this study is limited regarding the generalizability.
2.6. Conceptual Model 19
innovation both on a country level and an economic system level. More in detail we expect a variation in the slope of the regression line (regression coefficient) as well as in the intercept. In line with Akkermans, Castaldi and Los (2009) we expect better sector specific conditions for the IT sector in LMEs and slightly better conditions for manufacturing industries in CMEs. This leads us to the following hypotheses: Hypothesis 4a: IT companies from LMEs show a higher average level of R&D inten-sity (intercept) and a stronger effect (correlation coefficient) than competitors from CMEs.
Hypothesis 4b: Manufacturing companies from CMEs show a higher average level of R&D intensity (intercept) and a stronger effect (correlation coefficient) than com-petitors from LMEs.
Hypothesis 4c: The variation in sector-specific advantages depends more on country level than on the economic system.
2.6. Conceptual Model
The proposed research is an inductive research. We will use Interval or Ratio-scaled data in order to test our proposed hypotheses in a quantitative way. Figure 2.6 illustrates our expectations. Based on the previous research we expect a non-linear relationship, more specifically an inverse U-shaped relation between the independent and dependent variables. The data collection is based on secondary data from the Orbis database. The following conceptual model indicates the further procedure of the thesis.
After describing the conceptual model and the hypotheses that guide our research, the following chapter presents data and methods which have been used. First, we explain the data and the sample as well as the sampling design. Next, we will give a comprehensive overview of the measurements applied for the dependent and independent variables. To finish this chapter we will provide a detailed description of our methodological approach as well as of our statistical model.
3.1. Data and Sample Design
When compiling the sample we collected corporation information mainly from the Orbis database. This study considers only public listed companies because of disclo-sure obligations. Due to these disclodisclo-sure obligations we endisclo-sure a consistent reporting standard regarding released financial data. Furthermore, we used four criteria to de-termine the Orbis1 output, which are listed in the following table:
Sample criteria
1 Regional criteria United States of America, Germany, Canada, Nether-lands, United Kingdom and Japan
2 Industry criteria NACE Rev. 2 (Primary codes only): C- Manufactur-ing and J - Information and communication
3 Category of firms Very large companies, Large companies, Publicly listed companies and Formerly publicly listed com-panies
4 Years available 2014, 2013, 2012, 2011, 2010
Table 3.1.: Sample criteria Source: own model
First, we look only at companies from the USA, UK and Canada. These countries represent the LMEs in our sample. In contrast, Netherlands, Germany and Japan describe the CMEs part in this study. We selected these countries for two reasons. First, most of the criticism regarding innovation in the VoC approach refers to the limited variation in the country selection (Taylor, 2004; Akkermans, Castaldi and Los, 2009). Second, most of the previous studies focused on US based firms or on
3.1. Data and Sample Design 21
specific countries. Thus, the generalizability is still limited, why it seems useful to analyse multiple countries (Kirca et al., 2011). As mentioned, we follow the grouping of CMEs and LMEs counties of Hall and Soskice (2001). However, we recognized strong criticism regarding this specific composition in the context of innovation (Tay-lor, 2004). Akkermans, Castaldi and Los (2009) determine three main reasons which we consider in our data collection. (1) The evidence and comparison of only two countries is not enough to support the hypothesis of Hall and Soskice regarding LMEs and CMEs. A larger group of countries is needed to draw a significant con-clusion on a group of countries. Therefore we refer to a broader group of countries, as mentioned. (2) The different industry effect of radical and incremental innovation is not fully agreed yet. To gain more detailed insights we use companies from the IT sector to represent radical innovation industry and manufacturing industry for incremental innovation. (3) Scholars argue that the technology life cycle changes the type of innovation over time. Akkermans, Castaldi and Los (2009) argue that radical innovation occurs more often in the early stages of a technology than dur-ing an incremental innovation. Thus, the detractors argue that innovation is more dynamic over time than stable, as assumed by Hall and Soskice. With a multilevel approach we will gain new insights which can be used as a starting point for further research towards a dynamic approach. In chapter 6 we will emphasise the dynamic effects in detail more.
Second, in this composition, the industry criteria were chosen, on the one hand, because of the relevance of findings for these industries. On the other hand, because of our sample being more heterogeneous than previous studies. Thus, we will gain more insights into the interplay, not only regarding the industry specification but also in regard of the economic system, which was often criticised in the VoC approach of Hall and Soskice in prior studies (Taylor, 2004).
Third, determination of the category of the firms is based on two advantages. (1) the fact of having only large and very large firms helped to ensure that the sample firms had adequate size to achieve the economies we hypothesized. In return, for smaller firms the availability of data is small and incomplete, especially regarding the geographically segmented data or product-line specific financials. The inclusion of smaller firms would have resulted in a significant lack of data in our sample, which in turn would have been at the expense of representativeness.
3.2. Dependent Variable
The dependent variable in this research is firm performance. In literature different approaches are used for measuring firm performance. In general it is important to distinguish between accounting based measures and market-based measures (Lu and Beamish, 2004). According to Hitt, Hoskisson and Kim (1997) three accounting-based measures were initially considered as possible indicators of firm performance: (1) return on assets (ROA), (2) return on sales (ROS), and (3) return on equity (ROE). Whereas, ROS and ROA generated similar findings in their model and were highly correlated (r=.91), Tallman and Li (1996) use ROS as dependent variable and provide an extensive argument for employing sale-based measures to avoid the effect of differential assets valuations resulting from investment and/or depreciation. Fur-thermore, EBITDA was also used by other researchers (Wan and Hoskisson, 2003). Another frequently used market-based measure is Tobin’s Q, a ratio defined as the market value of assets divided by the replacement value of assets (Lu and Beamish, 2004). Due to limited data access we will use an accounting-based measurement, namely ROA, which is in line with prior studies (Tallman and Li, 1996; Lu and Beamish, 2004). Also, Wan and Hoskisson (2003) state that ROS is also a suit-able predictor. ROA is calculated by the ratio of net income to total assets. ROA demonstrates how efficiently assets are used in management to generate earnings. 3.3. Research and Development Intensity
The independent variable R&D intensity is highly related to innovation measure-ment. Often used measurements in this respect are the number of patents, (Wieser, 2005) or new product introduction (Hitt et al., 1996). According to Hitt, Hoskisson and Kim (1997), the ratio between R&D expenditures and a firm’s total number of employees is an appropriate ratio for the measurement. Hitt, Hoskisson and Kim argue that this ratio avoids problems of an artificial relationship to firm size, in the present case related to sales. Other studies use the ratio of R&D expenses to total sales (Gentry and Shen, 2013; Lu and Beamish, 2004). We follow the approach of Hitt, Hoskisson and Kim (1997) and measure the R&D intensity by the ratio of R&D expenses to the number of employees. We compare this ratio with the ratio of R&D expenses versus operating revenue. The latter shows more reliable results. Moreover, due to missing employee data we used R&D expenses / operation revenue as independent variable.
3.4. International Diversification
3.5. Economic system 23
measurements in order to increase the validity, in other words, to make sure that the chosen approach measures what it is supposed to measure. The first measure-ment takes the weight of each global market region into account, what was mostly neglected in previous research. Moreover, this study is possibly the first, which in-vestigates this effect by applying this approach both to a more diverse sample and to the IT industry. Moreover, the prior measurements do not respect the relative importance of each global market region with respect to total sales of a firm. We follow the Hitt, Hoskisson and Kim (1997) approach. It is the entropy measure of international diversification which is defined as:
International Diversif ication (ID) =X
i P (ID)i ∗ ln 1 P (ID)i P (ID)i =
Sales per region T otal sales
In this formula P (ID)irepresent the sales attributed to global market region diverted
by the total sales of the company i. ln(1/Pi) is the weight given to each global market
region, or the natural logarithm of the inverse of the firms sales.
In the existing literature, the second measurement, the Herfindahl index is often used for determination of diversity. The index determines the number of geographic markets as well the proportion of value that is derived from each market region. We used a modified index. Regarding the modification we followed Lampel and Giachetti (2013). Thereby we used the P(ID) as the percentage of sales per region to total sales.
International Diversif ication (ID) = 1 −XP (ID)i2
P (ID)i =
Sales per region T otal sales
The main difference between the first and the second measurement is the relative importance of each foreign market. This is why we expected slightly different results. However, the results show no significantly different effect. Thus, we only considered the first measurement for all our calculations.
3.5. Economic system
section 1 figure 1.1. We will follow the selection of Hall and Soskice (2001) regarding LMEs and CMEs. However, we will take into account more countries for each group as explained in section 3.1. Therefore, we used country identification codes in our sample. By this procedure we create the basis for a multilevel analysis.
3.6. Control Variables
Control variables are firm size, financial leverage, product diversification, industry, country of origin (Headquarter). Firm size is a frequently used control variable related to diversity levels. Firm size is measured by the natural logarithm of net sales. Financial leverage is measured by debt-to-equity ratio (Total Liabilities / Shareholders’ Equity). Prior research has shown that industry effects have important impacts on cross-sectional variation of firm performance (Hitt, Hoskisson and Kim, 1997; Wan and Hoskisson, 2003). Thus, we used industry dummy variables to control these industry differences. Furthermore we use country dummy variables in order to control for country specific differences. Following the prior studies of Lu and Beamish (2004), we also use product diversification as a control variable. We compute product diversification as a Herfindahl measure. Product diversification will be calculated as follows:
P roduct Diversif ication (P D) = 1 −XP (P D)i2
P (P D)i =
Sales per category T otal sales 3.7. Statistical model
To test the hypotheses the following model with one dependent variable (Y) and two independent variables XID and XR&D as well as control variables will be utilized.
The following represents first our basic assumption. Thereon, we build the formula towards a multilevel regression model. The following equation 3.1 represents the basic model. i representing the company, and the squared terms representing the non-linear effects.
Yij = α0+ β1XID2 i+ β2X 2
R&Di+ controls + ε (3.1)
In the next equation 3.2 the moderating effect has been added both for the linear and for the non-linear effects.
3.8. Modelling Procedure 25
In model 3.3 we include the multilevel adjustments, we add the higher levels, see figure 1.1 in section 1. The multilevel effect will be tested only in a linear manner and the dependent variable change to R&D which represents innovation. We expect that both intercept and slopes, vary around the overall model. First, we include a random intercept. What we do, is adding a component to the intercept that measures the variability in intercepts, in our model represented by ν0j. Thus, the
intercept changes from α to α + ν0j.
Next, we include the adjustment, ν1j for the random slope for the effect of home
country on industry. Therefore, the gradient changes from β1 to β1 + ν1j. The j in
the model represents the levels of the variable over which the intercept varies. The following model 3.3 represents all adjustments for an comprehensive overview.
Yij = (α0 + ν0j) + (β1+ ν1j)XITi+ controls + ε (3.3)
In order to simplify the model and link it more directly to a simple linear regression model, we take out some of these extra terms. Thus, we can declare this model to be a basic linear model, provided that we have replaced the fixed intercept and slope (α0 and β1) by their random counterparts (α0j and β1j). This leads us to following
model 3.4: Yij = α0j + β1jXITi+ controls + ε (3.4) where α0j = α0+ ν0j β1j = β1+ ν1j 3.8. Modelling Procedure
After collection the data from Orbis we further processed the data in Microsoft Excel, using the criteria, depicted in Appendix D table D.2. Based on these steps we generated our final sample which is used for all our analyses.
different reporting standards or lacking information from the Orbis database itself. As already mentioned there is a large inequality regarding the amount of companies from CME compared to LME. We expected to find a higher presence of LME com-panies because of the U.S., this assumption being based on population differences. The US population is around 319 million and is larger than Japan, Germany and the Netherlands together which amount to a total population of 223 million (World Bank, 2015b). We therefore expected a higher presence of firms from LMEs in our sample. This in turn also indicates lacking information from Orbis. Furthermore the large elimination in step 4, see table D.2 in Appendix D, provides more evidence supporting our assumption. 60% of the total sample has been excluded in this step because of missing regional data.
3.9. Estimation Method
3.10. Evaluation of Method Assumptions 27
3.10. Evaluation of Method Assumptions
In order to provide the the best linear unbiased estimation (BLUE) we refer to four crucial assumptions. The following will show that our dataset satisfies these assump-tions. Before we tested these assumptions we conducted two pre analyses. Fist, we analysed our dataset regarding missing values. The result shows that no missing data (0.00%) are existing in our sample. This can be explained by our modelling proce-dure as described in table D.2. The second analysis considered outliers in the sample. We used the Mahalanodbis measure (D2) which determines the distance of a certain
We assume BLUE conditions as described in the previous section. In this chapter first we explain the importance of the applied tests and present results of these four assumptions. Next, we will test our hypotheses and explain the results in detail.
4.1. Homoscedasicity
Ordinary least squares (OLS) models are based on the assumption that at each level of predictor variables the variance is constant, homoscedasticity (Field, 2009). Is this assumption not violated, which means that the error variance for all observations is unequal we assume heteroscedasticity is present. In case of heteroscedasticity we ob-tain biased estimates of standard errors. This in turn implies, that the test statistic is biased as well. Thus, it would lead to incorrect hypothesis tests, confidence in-tervals and p-values. However heteroscedasticity itself does not automatically imply that the coefficients are biased. The OLS is still linear and an unbiased estimator. According to Hill, Griffiths and Lim (2012) it implies that it is no longer the the best approach with the smallest variance. Since the OLS minimizes the sum of squared errors, the main problem of heteroscedasticity is that observations in connection with potential errors are more weighted than others. Thereby the linear regression becomes adulterated from the true regression line. The figure E.3 in Appendix E shows that the results are randomly and evenly dispersed throughout the plot. This in turn is indicating that the assumptions of homoscedasticity have been met (Field, 2009).
4.2. Endogeneity
The next condition states that the error terms of the variables are independent. If error terms are correlated they are either autocorrelated or serially correlated. In statical terms this means that the error terms are assumed to be uncorrelated and homoscedastic. If this assumption is violated in such as an independent variable (xi) is correlated with unmeasured variables (ei), the effect from the independent
4.3. Multicollinearity 29
ideal level being around two. However the Durbin-Whatson test depends on the order of the sample and is therefore not a appropriated measurement in our context. This is also in line with Keller (2012) who states that the method is appropriate for time series data but not for studies like ours. Therefore the second analysis is a more reliable method in our context. The scatter plot in Appendix E, shows that the standardised predictors on the x-axis and the standardised residuals values on the y-axis show that the results are almost meeting the assumption. The results should be in the range between +3 and -3 on both axes (Field, 2009). In our case we have only 10 results which are out of range. In other words only 1.9% of the residuals are outliers. This pattern is indicative for the fact that the assumptions of independent error terms have been met. Thus we can assume that our sample is slightly biased, but it is not a threat to our model.
4.3. Multicollinearity
Multicollinearity is not only critical for OLS, it is also a key assumption in multilevel models. We refer to multicollinearity, also called collinearity or inter-correlation, subsist when one or more predictors are strongly correlated with each other (Keller, 2012). Perfect collinearity means that at least one predictor variable is in perfect linear combination with the others (i.e. if two predictor variables have a correlation coefficient of 1). In the case of perfect collinearity it becomes impossible to generate unique estimates of the regression coefficients. This is caused by an infinite number of combinations of coefficients which would work identically well (Field, 2009). In other words the values of b for each variable are interchangeable. According to Keller (2012) there are two consequences of multicollinearity. First, the variability of the coefficients is large. This implies that the coefficient is probably far away from the actual population parameter. Thereby there is a possibility of the parameters having opposite signs. Second, In case of small t -statistics we can assume that there is a no linear relation between independent and dependent variable (Keller, 2012).
In order to test the presence of multicollinearity we calculated the variance inflation factor (VIF). VIF is an index which determines the level of variance of an estimated regression coefficient which is caused by intercorrelation. There are no hard rules about critical VIF values. However, there are some generally accepted level of VIF values. VIF ”value of 10 is a good value at which to worry.” (Field, 2009, p. 224) Moreover, the tolerance statistic, which is calculated by 1/VIF provides more evi-dence regarding Multicollinearity. Values below 0.1 indicate serious problems and values below 0.2 are worthy to concern (Field, 2009).
average score of 0.867 and all values are above 0.7 which means we are far away from the critical factor of 0.2. Thus, we conclude that we have a small amount of multicollinearity, which is unavoidable (Keller, 2012). However the low level of collinearity poses little threat to our model. Based on these results we can say that our model is not biased by multicollinearity.
The results are reported in the following table 4.1. Table 4.1 shows the VIF per variable and the tolerance statistic. We followed the approach of Field (2009).
Test for Multicollinearity
Variable VIF 1/VIF
Firm size 1,086 0.920 Financial leverage 1,043 0.958 Product diversification 1,283 0.779 Netherlands 1,072 0.933 United States of America 1,214 0.824 Germany 1.147 0.871 United Kingdom 1.156 0.865 Manufacturing 1,114 0.897 Research and Development 1.165 0.858 International diversification 1.296 0.771
average 1.153
Table 4.1.: Multicollinearity Source: own results
4.4. Normality
The normality test is essentially important in the context of OLS and for multilevel models. First, the assumption of normality is that the values of the error term are normally distributed among their means. If the normality assumption is violated, this can lead to biased p-values. Thereby it affects the significance (Hill, Griffiths and Lim, 2012). We therefore tested our sample regarding kurtosis and skewness. The following table shows the results of this analysis.
Test for Normality
Variable Skewness Kurtosis Firm performance -1.968 11.07 Firm size 10.50 138.16 Financial leverage 9.19 147.41 Product diversification .64 -.29 Research and Development 5.83 47.84 International diversification .81 .04
4.5. Results of Hypotheses 31
Based on these results we corrected the skewness and kurtosis of our control variables, firm size and financial leverage using natural logarithm. We also tested our sample with corrected firm performance variables. However, the results did not change significantly only the effect became weaker. The Skewness and Kurtosis here can be explained by the nature of our sampling procedure. We selected only large and very large companies, thus excluding an important part of the normal distribution. We decided to use the original data and to continue without an adjustment in this case. Furthermore, the same holds for R&D. However, we are aware that our dataset is not perfectly distributed and thereby the p-values can be slightly biased.
4.5. Results of Hypotheses
The means and standard deviation as well as the correlation for the variables used for this analyses are presented in table 4.3.
The following analysis is based on a two stage approach. In the first stage we observe the Hypotheses 1 to 3 by applying OLS as well as curvilinear regression analysis. In the second stage, we test Hypothesis 4 by using a hierarchical linear regression model.
ROA is the dependent variable for the six models in the first stage and in the mul-tilevel analysis. We report the results of the first stage in table 4.4. Model 1 is the baseline model, it includes only the control variables. Product diversification has a strong negative effect, indicating that firms’ performance decreases with each addi-tional unit of product diversification. Firm size had a significant positive impact. Financial leverage also has a significant but negative effect. As for the Dummy vari-ables, we excluded Japan for the country Dummy and IT for the industry Dummy. Netherlands, Germany, US and the United Kingdom all have a positive influence if compared to the excluded country dummy, Japan. Thereby Germany and the UK have significant effects. Moreover manufacturing companies have a higher impact on performance than IT companies.
We tested Hypothesis 1 in models 2 and 3. In these models we built up the test for the inverse U-shape relationship between R&D intensity and firm performance in two steps. First, we tested the linear effect in Model 2. Secondly we tested the squared term in Model 3. Model 2 shows a significant negative effect. In Model 3 the squared term has a significant negative effect and the linear effect was positive in Model 3. Thus we can say an inverse U-shape relationship is existing. The change in the F-statistic (4F= 3.35, p < 0.01) shows that including the squared term significantly improved the model fit. Furthermore the change in the R2-(4R2=.06) and adjusted R2-(4R2
adj=.05) values indicates that the explanatory power has increased. Thus,
In order to test Hypothesis 2 we used Model 4 and 5. We followed the same procedure as for the first Hypothesis. In Model 4 we tested the linear effect finding a linear negative effect but the result was not significant. In Model 5 we tested the squared term. The results predict a stronger inverse U-shaped (2.72 to -2.27) relationship than for the first hypothesis, but they are not significant. Furthermore the F-statistic change (4F= -0.8, p < 0.01) indicates that the model does not fit the data. Moreover the change in the R2-(4R2=.01) and adjusted R2-(4R2adj=.00) indicates that the explanatory power increases only marginally for R2 but not for Radj2 . Thus, the Hypothesis 2 is not fully supported, the results indicate the expected effect, however, without significance.
The moderating effect between R&D intensity and international diversification which predicts Hypothesis 3 is tested in Model 6 and 7. In Model 6 we tested the interaction between the two independent variables in a linear manner. The results show that there is a slightly negative moderation but without significance. In Model 7 we tested the moderation effect with the squared R&D and international diversification val-ues. The results show highly significant positive moderation. The F-statistic change (4F= 2.44, p < 0.01) shows that the inclusion of the squared term significantly improved the model fit. Furthermore R2-(4R2=.08) and adjusted R2-(4R2
adj=.08)
values changed. This in turn indicates that the explanatory power increased. Con-sequently, we can say that Hypothesis 3 is supported.
The following figure summarizes and illustrates our findings, i.e. the non-linear effect between international diversification and the positive moderation effect of R&D intensity.1 The graph shows only positive effects.
Figure 4.1.: Moderation Effect of R&D intensity on the relation between ID and ROA, Source: own results
4.6. Results of Multilevel approach 35
4.6. Results of Multilevel approach
In order to test the fourth Hypothesis we first introduce our model step by step. First, we will consider the R&D intensity for each firm and therefore calculate a mean (2009-2014), as indicated by red points in the following figure 4.2. Next we calculate a sample mean for each country, indicated by the green Gauss normal distribution curve. The R&D intensity (our observations from the first step) vary within a country. In order to take this variation into account we build new country means as indicated. Likewise the countries are nested within economic systems. Therefore we calculate the mean for each economic system. Thus we can consider each of the country means to vary towards the overall economical system mean. Finally, we can calculate the overall grand mean. To describe this from a mathematical point of view we introduce the variance component model in the following figure 4.2 which illustrates the equation:
Figure 4.2.: Variance components model, Source: Huber (2013)
Yijk = µ + µi+ µij+ eijk
where, i = Economic system, j = Country, k = Observation(R&D)
The observation Yijkis a mixture of fixed effects represented by µ and a random part
of the model represented by eijk (residuals). So we have broken the residual part into
three different components. First we can calculate the deviation of the country mean to the grand mean µi (Blue). Next we can calculate the deviation and the residuals
with respect to the difference between the economic system mean and country mean µij (green). Finally we end up with some residuals (random part). Based on that we
intensity of a firm. Therefore we made certain assumptions: µi ∼ N (0, γ2)
µij ∼ N (0, τ2)
µij ∼ N (0, σ2)
Kendall tau rank correlation (τ ) and Kruskal’s gamma coefficient (γ) are recom-mended measures in order to measure the association between two quantities (Field, 2009). The results of both test show significant (p < .000) evidence that the two levels depend on each other and therefore a multilevel model is needed. Further-more we already demonstrated by the VIF results that our model poses very low multicollinearity. The last assumption will not be tested again because we already examined this issue in section 4.4 and showed that normality is not threatening for our model.
Before we start testing Hypothesis 4 we examine whether the use of multilevel mod-elling is statistically supported or not. According to Huber (2013) the total variance should be larger than 10% between the observation and the first level. In order to test the so called nonmodel or unconditional model, we created an empty model which is also the baseline model (Garson, 2013). This implies running the model without the predictor variables. The so called interclass correlation (ICC) deter-mines the amount of the total variance which is dedicated to the upper level. The result provides us variance estimation at level 2. (i.e. companies at level 1 and coun-tries at level 2). The result shows that 17.86% of the variation in R&D intensity are owed to country variation. This in turn is above the minimum value of 10%. Thus we can say that further multilevel modelling is necessary. The chi square LR test was 40.13, df=02 and thus highly significant (p < 0.000). If the ICC value had been below 10% a OLS regression would have been better. However, this is not the case and the model is highly significant. Thus we can test the last hypothesis by using a nested hierarchical linear model (ESS EduNet, 2015).
First, we are going to test the first part of the hypothesis 4a/b, higher intercept for IT firms with a LME background and higher intercepts for Manufacturing firms from CMEs.
