Science fiction as a business fact.

(1)

Science fiction as a business fact.

The relationship between automation and firms’

international expansion behaviour

Master thesis IB&M

Economics and Business Faculty University of Groningen

Final version: January 22nd, 2018 Word count: 14,951

Author: Supervisor:

Harrasser, Elfi Dr. R. W. de Vries

Student number: S3240509

Leopoldstraße 206, 80804 Munich Co-assessor:

(2)

ABSTRACT

This paper studies the relationship between advances in automation and international expansion behaviour of American manufacturing firms. As today’s economic environment is changing rapidly, firms are constantly under pressure to develop new business strategies in order to survive and stay competitive. Thereby, internationalisation can be a viable and profitable growth strategy. However, based on the assumption that firms experience severe difficulties and liability of foreignness when expanding into new geographic locations, the following study hypothesises that automation benefits reduce firms’ pressure to go abroad. In order to test the hypothesis, the study executes partial correlation analysis based on time-series data of 199 U.S. manufacturing firms from the period 2014-2016. Moreover, it develops a novel quantitative approach to evaluate firms’ level of automation and contributes to the field of strategic management. Findings indicate that whilst automation can be an alternative for foreign production, it only plays a subordinate role for firms’ internationalisation pace and geographic dispersion. However, no relation was found between firms’ increasing investments in automated machinery and foreign sales. These findings come with some implications for theory, practice and future research.

(3)

ACKNOWLEDGEMENTS

After an intensive period of four month, today is the day to look back and express my personal thanks to the people who have supported and helped me so much throughout this last section of my studies.

First and foremost, I offer my sincerest gratitude to my supervisor, Dr. Rudi de Vries, who gave me the opportunity to engage with a topic that I was interested in since the beginning of my Master’s studies. I am thankful for his guidance, constructive criticism and valuable advice, but also for giving me the chance to make this paper my own work.

In addition, special mention goes to my fellow students from the IB&M intake 2016/2017, who accompanied me on the adventurous journey through Groningen’s university- and student life. Without your inspiration, motivation and friendship, my Master’s time would not have been the same.

Finally, but by no means least, special thanks go to my family and my boyfriend for almost unbelievable support and confidence. No matter which academic path I took, you were always on my side and encouraged me to follow my dreams and passion. I am extremely grateful for your love, prayers, caring and sacrifices for educating and preparing me for my future.

Thank you very much,

(4)

LIST OF FIGURES

Figure 1: Model of international market entry. ... 8

Figure 2: The four elements of automation ... 10

Figure 3: Factors influencing firms’ automation implementation ... 11

Figure 4: Conceptual model ... 17

Figure 5: Final sample. Distribution of NAICS 2017 ... 20

(6)

LIST OF TABLES

Table 1: Internationalisation theories ... 4

Table 2: Variables, measures and sources of data ... 27

Table 3: Normality Test ... 30

Table 4: Descriptive Statistics for the time period observed ... 32

Table 5: Automation Statistics per year ... 33

Table 6: Internationalisation Statistics per year ... 33

(7)

LIST OF ABBREVIATIONS

AI Artificial Intelligence

CEBR Centre of Economics and Business Research

IFR International Federation of Robotics

ILO International Labour Organisation

NAFTA North American Free Trade Agreement

NAICS North American Industry Classification Systems

OLI Ownership, Location and Internalisation

RIA Robotic Industries Association

SEC Securities and Exchange Commission

U.S. United States

(8)

1. INTRODUCTION

Internationalisation drastically changed the way firms are doing business nowadays: The continuous omission of economic borders does not only influence the nature of firms’ strategic behaviour, but also the exploration of competitive advantage as well as the boundaries of inter-firm competition (Bartlett & Ghoshal, 1989; Sanders & Carpenter, 1998). To remain competitive in the global market place, firms are increasingly diversifying their activities across national borders into new geographic locations (Hitt, Hoskisson & Kim, 1997). Many of them try to reduce market imperfections by outsourcing predictable, not very complex, but labour intensive activities to low-cost countries (Molnar, Pain & Taglioni, 2007) and realizing economies of scale (Dunning, 1988). By doing so, most firms are able to raise both efficiency (Franko, 1989) and profitability (Vermeulen & Barkema, 2002). At the same time, operating in multiple countries also implies dealing with high levels of complexity that derive from heterogeneous cultural, institutional and competitive environments. Cultural differences, for instance, can lead to less efficient knowledge sharing and synergy formation among business units, which consequently hampers both exploitation and exploration of competitive advantages (Gomez-Meija & Palich, 1997). If firms do not have the ability to manage and overcome such liabilities of foreignness (Zaheer, 1995), high additional costs are the consequence (Lu & Beamish, 2004; Tallman & Li, 1996). Research therefore assumes that many firms base their international market selection on geographic and institutional proximity to avoid possible difficulties (Rugman, 2000) and hence are no real international market players (Rugman & Hodgetts, 2001).

This notion is especially interesting when considering firms’ recent internationalisation strategies in the world’s largest economy, the United States (U.S.)1. In particular, the reshoring of U.S. manufacturing firms has attracted considerable attention over the last years. News regularly reported on firms building production facilities domestically and shifting operations back to the U.S. since 2009 (Crooks, 2012; Loeb, 2013; Vlasic, 2017). General Electric, for instance, moved part of its production from China and Mexico to Kentucky, New York and Ohio in 2009 and explained its decision with rising labour- and shipping costs abroad (The Economist, 2013). In 2015, it was the automotive manufacturer Ford who announced the shifting of its most sophisticated small engine production from Spain and Mexico to Cleveland (Priddle, 2015). Interestingly, this general development has been further confirmed by an online

(9)

survey conducted in 2012, which revealed that around 40% of U.S. manufacturing firms are actively planning to translocate production back to the U.S. (Sirkin, Zinser, Hohner & Rose, 2012).

Alongside this development in global economy, another business strategy has gained significant importance over the last years: Industrial automation describes the implementation of modern techniques and practices in manufacturing (Acharya, Sharma & Gupta, 2017; Martin, Ulich & Warnecke, 1991) to achieve sustainable competitive advantage and enhance business performance (Gunter & Butler, 1999). In the past three decades, industrial organisations have radically changed their manufacturing processes through the adoption of automated manufacturing techniques (Dean, Yoon & Susman, 1992). Research shows that firms continuously investing in industrial automation can gain considerable strategic benefits, including increased market share and cost reduction, improvement in strategic performance and return on investment, as well as improved productivity (Gunter & Butler, 1999).

When looking at the U.S. economy, a steep improvement in firms’ exploitation of innovative technologies can be observed: According to a lately published study by the Centre of Economics and Business Research (CEBR)2_{, U.S. investments in automated machinery and}

processes increased by 30% between 2011 and 2015. Moreover, the American sector association RIA (Robotic Industries Association)3 reports that U.S. manufacturing firms were ordering over 6,600 industrial robots within the first three months of 2017, which equals an increase of 20% compared to the same period last year. This development can be mainly explained by decreasing robot prices, which, in contrast to the cost of labour, have fallen by 40 to 50% since 1990 and will further drop another 22% by 2025 (Sirkin, Zinser & Rose, 2015). Observations further reveal cost savings of about 70 to 80% compared to the cost of conventional production factors: Whilst firms in electronics manufacturing, for instance, pay $24 per hour for an average employee to perform routine tasks, they only spend $4 when using a robot which is able to execute the same tasks in even less time (Sirikin et al., 2015). As a consequence, firms increasingly view automation as the economically better bet (Frey & Osborne, 2016).

2_{The CEBR is a leading economic consultancy in the UK and supplies public, government and}

professional bodies with independent economic analysis and forecasting.

3_{The Robotic Industries Association (RIA) is an American}_{trade group specifically organised to serve the}

(10)

Although industrial automation has received extensive attention in research, academic studies predominantly examine the effect of automation implementation on either operational performance (Carlsson, 2012; Voss, 1988) or employment (Frey et al., 2016; Hudson, 1982; Manyika, Chui, Miremadi, Bughin, George, Willmott, & Dewhurst, 2017). Thus, another interesting thematic field, namely the simultaneous occurrence of the latter two phenomena, remains, to my knowledge, largely unexplored. Economic observers already suspect a relationship between automation and firms’ international expansion behaviour (c.f. The Economist, 2013). However, there is still no empirical evidence if investments in automation can possibly substitute for firms’ low-skilled, cheap workforce requirements abroad and consequently reduce the pressure to internationalise in order to stay competitive. Moreover, the relationship between automation and observable contemporary regionalisation tendencies (Rugman & Verbeke, 2004), which lead to lower liability of foreignness and thus to substantial cost savings, has also been neglected by research so far. To bridge this gap, the present study assesses the relationship between industrial automation and international expansion behaviour of manufacturing firms in the U.S. Hence, the following research question can be stated:

“Is there a relationship between advancing automation and international expansion behaviour of manufacturing firms in the United States?”

The study addresses this question by using time-series data (2014-2016) on automation and firms’ international expansion behaviour of a sample of 199 U.S. manufacturing firms. Moreover, it develops a novel quantitative approach to evaluate firms’ level of automation and gain insights into recent strategic developments. It is expected that automation plays an important role for firms’ growth strategies. Furthermore, this study assumes that advances in robotics and machine learning reduce internationalisation pressures firms experience when trying to stay competitive by simultaneously strengthening their strategic position.

(11)

2. LITERATURE REVIEW

2.1. Internationalisation as a main driver of economic growth

In today’s complex and constantly changing business environment, firms need to develop new strategies to successfully meet future challenges and benefit from emerging business opportunities. According to D’Souza and McDougall (1989), the ability to engage in foreign business can thereby constitute an important and inevitable step to ensure the survival and growth of firms. There are different ways firms can intensify their international operations, varying from the establishment of foreign subsidiaries to the development of international joint ventures and licensing agreements (Johanson & Vahlne, 1990). Moreover, motives range from resource and efficiency seeking (Dunning, 1988) to the exploration of knowledge and sustainable competitive advantage (e.g. Casillas, Moreno, Acedo, Gallego & Ramos, 2009). Research further shows that most manufacturing firms employ internationalisation strategies as a strategic reaction to increased competitive pressures (Cavusgil, Yaprak & Yeoh, 2008; Swamidass, 1993). Hence, international expansion can be seen as a viable growth strategy (McDougall & Oviatt, 2000) that positively affects manufacturing firms’ financial and strategic performance (Kotabe & Murray, 1996).

Considering current literature, there is no general agreement on the factors influencing firms’ internationalisation choice. However, there are several approaches trying to explain why firms decide to expand into foreign markets as well as how the internationalisation process looks like. As shown in table 1, one can generally distinguish between four main theoretical streams of internationalisation (Whitelock, 2002): (1) the Eclectic Paradigm and transaction cost analysis; (2) the Uppsala model of internationalisation; (3) the interactive network approach; and (4) the business strategy approach.

Table 1: Internationalisation theories

Theory Influential feature Focus

Eclectic Paradigm & Transaction Cost Theory (Dunning, 1988; Coase, 1937)

Cost of transactions Firm

Uppsala Model

(Johanson & Vahlne, 1997)

(12)

Industrial networks & revised Uppsala Model (Johanson & Mattsson, 1986; Johanson & Vahlne, 2009)

Interaction Firm, customer,

competitor, supplier within market environment

Business Strategy

(Welford & Prescott, 1994; Reid, 1983)

Opportunity Resources Managerial philosophy Market Firm Firm

(1) Economic internationalisation perspectives, including Eclectic Paradigm and transaction cost analysis, argue that international market entry decisions are made in a rational manner, viewing the firm as a sensible entity that is able to choose between alternatives (Whitelock, 2002). According to Dunning’s Eclectic Paradigm theory (1988), the decision to expand into foreign markets and engage in foreign production is based on three dimensions: Ownership, Location and Internalisation (OLI) advantages. Consequently, firms’ competitive advantages can derive from benefits specific to the nationality and nature of their ownership, location advantages as well as benefits originating from internal production. Depending on varying degrees of OLI advantages, firms differ in their entry mode decisions (Dunning, 1988; 2000; 2001). In addition, transaction cost theory is a powerful tool explaining how firms evaluate internationalisation opportunities (Erramilli & Rao, 1993). Based on the assumption that markets are competitive and imperfect, the theory proposes that there are certain costs bound to market transactions, including coordination costs, contracting costs, search costs, controlling costs and reorganizing costs (Coase, 1937). According to Coviello and Martin (1999), firms’ location choices and organisational structures are influenced by the ultimate goal to minimize transaction costs. As a consequence, high transaction costs often result in firms’ decision to internalise transactions, whereas low transaction costs lead to market transactions and international expansion (Johanson et al., 1990).

(13)

internationalisation as a process in which firms gradually increase their international involvement. According to Johanson et al. (1977), exports are first realised through independent intermediaries, who are replaced by sales subsidiaries later on. However, firms’ ultimate goal is to manufacture within the foreign market. The process is dependent on firms’ constant commitment to translocate resources and activities abroad as well as on their continuous knowledge development about foreign markets and operations. In that regard, experiential knowledge, which can only be developed through personal experience, plays a key role in overcoming liability of foreignness. As a consequence, firms are first expanding into markets which are psychically close and operating in more distant markets later on as their experience and knowledge develop.

Although the theory has received substantial support (Bilkey & Tesar, 1977; Johanson et al., 1990), it has also been questioned on a number of counts: While some scholars stress firms’ “mixed” usage of internationalisation approaches (Buckely, Mirza & Sparkes, 1987; Turnball, 1987), others put the strict sequence of each stage into doubt (Root, 1987). Just as exporting through independent intermediaries is not always the initial market entry, producing abroad is not the ultimate aim for all firms in all markets (Whitelock, 2002). In response to extensive criticism, Johanson and Vahlne tried to adjust their theory several times. In 2009, they responded to recent developments and shifted their focus from firms’ autonomy whether to expand into foreign market or not to the emerging role of business networks (Johanson & Vahlne, 2009). Whilst the old model focuses on the concept of psychic distance, resulting in substantial liability of foreignness when expanding into distant markets, the revised Uppsala model addresses the importance of a firms’ position in relevant business networks. According to Johanson et al. (2009), firms are embedded in business networks, which can both foster and constrain international expansion abilities. High integration in such networks enables firms to develop their business abroad, whereas low involvement can hinder firms to exploit internationalisation opportunities. Consequently, the model’s focus shifts from liability of foreignness to liability of outsidership, relating to the extent a firm is integrated or not in such business networks. However, Johanson and Vahlne still value the importance of experiential knowledge, which can be partly embedded in business networks and helps firms to strengthen their network position.

(14)

networks of firms that interact on a regular basis and establish, develop and maintain lasting business relationships. “Each firm in the network has relationships with customers, distributors, suppliers… there are also important competitive relations’’, Turnball (1986; p. 305) argues. Consequently, four factors influencing the economic interaction process can be identified: the elements and processes of interaction; the atmosphere surrounding the interaction; the characteristics of the parties involved; and the environment within which the interaction takes place (Cunningham, 1986). Thus, firms do not only need to consider their own competitive position, but also the target market as well as important environmental factors.

Finally, the (4) business strategy approach focuses on firms’ pragmatism and stresses their ability to make trade-offs between the factors that might influence the internationalisation decision and the methodsthey adopt to realise it (Welford & Prescott, 1994). According to Reid (1983), international expansion is based on firms’ autonomy to choose between alternatives and „results from a choice among competing expansion strategies that are guided by the nature of the market opportunity, firm resources and managerial philosophy’’ (p. 51). In order to select the right market, Root (1987) introduces the influencing elements of market attractiveness, accessibility, psychic distance and informal barriers. Firms’ entry mode, on the other hand, will be influenced by both market and company characteristics „such as international trading history, size, export orientation and commitment'' (Turnbull & Ellwood, 1986; p. 168).

(15)

Figure 1: Model of international market entry. Source: Whitelock, 2002.

(16)

2.2. The new era of automation

Technological progress and innovation have significantly contributed to the increase of firms’ productivity, competitive position and strategic performance over the last hundred years (Stoneman, 1983). Nowadays, advances in robotics, machine learning and artificial intelligence (AI) are leading the global economy to a new era of automation, in which machinery is not only able to perform a range of routine physical work activities better and much cheaper than humans do, but also capable of executing activities that require cognitive capabilities (Nof, 2009). When looking at today’s business world, firms are intensifying their usage of robots and algorithms on production lines to manage inventory, collect data, optimize logistics and carry out other key operational functions (Manyika et al., 2017).

The concept of modern automation was first envisioned by Diebold (1952; p. 62-63) in his book about the automated factory: “If we could couple a group of production machines, or similar machines designed around the bundle of functions concept, by some form of inexpensive and flexible material handling equipment, and add a control mechanism to do the work normally done by the operator, we would have a factory completely automatic in terms of direct operation, although there would still be need for considerable indirect labour.“ At that time, mechanization was a common version of automation, referring to the use of powered machinery to help humans perform physical tasks. Technologies were mainly “deskilling” human labour as they simplified processes and tasks originally carried out by employees (Frey et al., 2016; Goldin & Katz, 1998). While mechanization represents the replacement of human power by machine power, modern automation adds the intelligence feature of automatic control to the process (Gupta & Arora, 2013; Nof, 2009). In reference to Brynjolfsson and McAfee (2011), automation is therefore no longer confined to routine tasks, but is also able to solve non-routine challenges by turning them into well-defined problems. Thereby, the access to large and complex datasets (big data), whose algorithms help to detect patterns and discover unexpected similarities better than humans can (Campbell-Kelly, 2009), becomes essential (cf. Plötz & Fink, 2009; Veres, Molnar, Lincoln & Morice, 2011). Consequently, modern automation can be defined as the process of following a predetermined sequence of activities without or little human intervention.

(17)

systems involved are platforms developed by humans to process a given range of activities in an independent and autonomous way. Furthermore, power source is needed (Nof, 2009).

Figure 2: The four elements of automation. Source: Nof, 2009.

When looking at different technologies and tools developed to enable automation, the rise of AI has attracted significant attention over the last decades (e.g. Bench-Capon & Dunne, 2007; Birnbaum, Flowers & McGuire, 1980; Dirican, 2015; Reiter, 1980). According to Nof (2009), AI is the ability of a machine system to perceive both expected and unexpected conditions, decide upon the actions that have to be performed and manage these actions accordingly. Consequently, AI research predominantly dedicates itself to the development of intelligent algorithms that are able to automate cognitive tasks (Frey et al., 2016). Key fields of AI study are machine learning, knowledge-based systems, language processing systems and computer sensory systems (Manyika et al., 2017). Moreover, robotics has been identified as an important subset of automation (Fiorini, Carbonera, Goncalves, Jorge, Rey, Haidegger, Abel, Redfield, Balakirsky, Ragavan, Li, Schlenoff & Prestes, 2015; Nof, 2009). In contrast to classical automation, robotics is focusing on the automation of motion and mobility and thereby often neglecting additional areas based on software, planning and optimization, decision-making and collaboration (Nof, 2009). However, new technological approaches like cloud robotics stress the importance of integrating cloud computing and industrial robotics for smart manufacturing environments (Yan, Hua, Wang, Wei & Imran, 2017). When interacting with the cloud, robots can overcome information and learning constraints by simply downloading information and communicating with other systems (Wang, Wan, Li & Zhang, 2016).

(18)

reprogramming can be very time consuming and difficult to manage (Nof, 2009). Conversely, technologies belonging to flexible automation can quickly incorporate product changes and induce new product lines. According to Wadhwa (2012; p. 444), flexible automation allows “rapid reconfigurability of the production system in order to manufacture several different products, achieving high degree of machine utilization, reduction of in-process inventory, as well as decrease in response times to meet the changing customer preference“. As a consequence, firms integrating flexible automated tools are able to better meet uncertainty as well as environmental changes (Sethi & Sethi, 1990). In reference to McDermott and Stock (1999), flexible automation further contains a promise often characterised as the “Holy Grail” of production: “the flexibility to produce customized products combined with the efficiency of mass production, together with a host of organisational benefits” (p. 522).

Therefore, besides preparing firms for economic change, the deployment of automation technologies can lead to a lot of additional advantages. According to Manyika et al. (2017), automated processes enable firms to improve their strategic performance by enhancing quality and speed, reducing errors, cutting labour and production costs and realising outcomes that go beyond human capacities. Furthermore, automation contributes to productivity and has been associated with improved safety, reduction of waste, higher customer satisfaction and reduced variability (Gunter et al., 1999; Gupta et al., 2013). This is in line with prior research, which focuses on the main purposes automation is implemented for. The first one is to ensure a more precise performance, followed by the requirement to keep the respective performance stable and less prone to error. Moreover, firms want to overcome the capacity limitations human have in order to increase speed, efficiency and safety (Hollnagel, 2003).

(19)

However, automation pace and extent differ a lot across manufacturing firms (Manyika et al., 2017). In reference to Acharya et al. (2017), this phenomenon can be best explained by the presence and significant impact of certain factors on automation implementation. As shown in figure 3, one can distinguish between three categories: On the one hand, environmental factors, such as government policies, vocational education, unemployment and labour unions, do not only affect firms’ operations and growth, but also how automation is perceived and consequently encouraged in society. Driving forces factors, on the other hand, represent internal and external factors that influence the decisions firms make to stay competitive, including management support, strategic planning, competitive priorities, database management and computer aided design. Finally, continued technical progress and effective usage of automated technologies form the third category of technological factors. These factors do not only influence how firms operate in the modern world of economy, but also which technological solutions are used to perform automated processes.

(20)

2.3. Automation’s increasing importance in firms’ internationalisation strategies

Advances in technology, global competition and high levels of uncertainty are key factors urging firms to the development of new businesses strategies in order to survive in highly competitive markets (Choe, New & Tong, 2015). In response to these increasing pressures, many firms identified geographic diversification as a viable growth strategy (McDougall et al., 2000). According to literature, one of the most powerful forces behind internationalisation is the existence of differing factor prices across national borders (Feenstra & Hanson, 1996; Kohler, 2004). Consequently, many manufacturing firms shift their focus from the whole production cycle to their core competencies, which mainly include more technologically advanced production stages (Hijzen, Görg & Hine, 2005). At the same time, low-skill, predictable physical activities are outsourced to developing countries in order to save expensive staff costs at home (Ekholm & Hakkala, 2006; Molnar et al., 2007). Thereby, firms can either decide to acquire goods from independent suppliers abroad instead of producing them itself, or relocate their own activities to low-income countries (Ekholm et al., 2006).

(21)

All additional costs firms are facing when doing business abroad can be subsumed in the concept of “liability of foreignness” (Hymer, 1976; Kindleberger, 1969; Zaheer, 1995). According to Zaheer (1995), the liability of foreignness can arise from the unfamiliarity of the environment, from political, cultural and economic differences, as well as from the need to coordinate and manage operations across national borders. Moreover, literature distinguishes between four dimensions that influence the costs associated with internationalisation (e.g. Ghemawat, 2001, 2003; Kostova, 1997; Zaheer, 1995): (1) Cultural distance, for instance, reveals itself in different languages, races, religions and social norms. It often leads to misleading communication and different understandings of organisational procedures, which result in higher costs. (2) Administrative and political distance, on the other hand, focuses on expensive barriers foreign markets can raise. It covers inter alia tariffs, administrative restrictions on foreign investments and preferences for domestic competitors. Furthermore, weak institutional settings can create depressions influencing international operations significantly. (3) Costs of transportation, travel, communication and coordination over distance and across time zones are affected by geographic distance. (4) The fourth dimension of economic distance can be best observed at the organisational level and manifests itself in the type of trade and partners firms choose.

As rising transaction costs and increasing internationalisation difficulties significantly influence firms’ competitiveness (Salmi, David, Blanco & Summers, 2016), firms are willing to adjust their strategic directions in order to survive and be successful. Thereby, automation can act as an essential means.

2.3.1. Saving costs and being more flexible

When considering the factors that influence industrial organisations’ survival and success the most, cost plays a key role (Layer, Brinke, Houten, Kals & Haasis, 2002). It does not only represent an important component of firms’ business strategies, but also has sufficient weight on business decisions and managerial policies (Niazi, Dai, Balbani & Seneviratne, 2006). Nevertheless, firms incur high additional costs when developing relationships with potential foreign business partners, sharing practices and guaranteeing well performing operational processes to stay competitive (Schmidt & Sofka, 2009).

(22)

manufacturing firms that consider cost reductions as the major benefit of investments in flexible automation (Farley, Kahn, Lehmann & Moore, 1987). Moreover, as the real cost related to computing and automation is currently declining, firms experience substantial economic incentives to replace labour for automated machinery (Frey et al., 2016). Advances in robotics and machine learning can further increase firms’ flexibility in terms of handling variations in demand, internal processes, product variations and environmental changes (Wadhwa, 2012). According to Gerwin (2015) and Wadhwa (2012), they are a viable answer to increasing economic and social uncertainties, whose influence on firms has escalated over the past 20 years. In addition, they can be an essential means of firms’ rising flexibility requirements in an ever more customer-driven manufacturing environment (Legat, Lamparter & Seitz, 2010). As a consequence, it can be assumed that the more firms focus on innovative technologies and the implementation of specialized equipment, the more independent they are from economic pressures and competitive dynamics. Investments in automation technologies enable firms to benefit from extensive cost reductions and improved productivity at home and reduce economic pressures that urged them to expand into unknown low-income countries in the past. This development might relieve many firms from the complexities of internationalisation and thus lead to significantly decreasing liability of foreignness. This study therefore expects that with the rise of automation, firms’ foreign production and manufacturing facilities will decrease. Moreover, it suspects a negative relationship between automation and firms’ international expansion activities:

Hypothesis 1: Automation is negatively related to firms’ foreign production.

Hypothesis 2: There is a negative relationship between automation and the number of

internationalisation activities firms execute in a given year.

2.3.2. Operating in well-known markets

(23)

This assumption is in line with observations that stress general regionalisation tendencies of firms (Rugman, 2000; Rugman et al., 2004). In reference to Rugman et al. (2001), large firms conduct most of their business activities in regional blocks, also defined as the “triad” markets of NAFTA, Asia and the European Union. Research suggests that this phenomenon can be explained by transaction cost economics (Rugman et al., 2004, 2005; Seno-Alday, 2009), stressing that most firms can deploy their firm specific advantages abroad only if they are willing to make high additional investments in order to adapt their operations to the foreign business environment. However, the latter adaption process is less needed in firms’ home regions, where they are facing similar institutional and economic environments (North, 1991) and are able to take advantage of economies of scale. In contrast, if firms expand beyond a quite homogeneous regional market, increased costs are the result (Rugman et al., 2004). Consequently, as firms try to avoid high liability of foreignness and challenges deriving from internationalisation (Zaheer, 1995), many of them replace their business activities to markets that are psychically close, namely to their regional blocks (Rugman et al., 2001). Research further indicates that firms realise business activities in more distant markets only after their knowledge and experience developed (Johanson et al., 1977). In addition, there is empirical evidence that industrial manufacturers locate their new production subsidiaries closer to existing ones to reduce cross-border spatial transaction costs (Jiang, Holburn & Beamish, 2016). Thereby, cultural, political, geographical and economic differences are less complex and easier to manage and control. Following these observations, the study assumes that automation can act as an essential means to reduce internationalisation pressures firms are normally exposed to by offering a viable and promising growth alternative. Consequently, firms are less required to expand their business activities into distant and unknown markets as they are able to save costs and improve performance in their home regions, too. Hence, the study expects that firms’ geographic dispersion diminishes as automation is on the rise. In order to investigate the relationship between automation and the number of countries firms expand to when internationalising, the following hypothesis will be stated:

(24)

2.3.3. Increasing sales and strengthening firms’ strategic position

However, there is no empirical evidence that automation benefits hamper firms’ exporting behaviour. On the contrary, research found that decreases in transaction costs, which can also derive from proceeding automation tendencies, can increase firms’ market share in foreign markets (Yeaple, 2005). Anwar and Sun (2017) further emphasise the positive impact of firms’ rising productivity on both exports and product quality. Moreover, the implementation of automated machinery and processes can help firms to achieve competitive advantage and hence strengthens firms’ strategic position (Manyika et al., 2017). As a consequence, this study assumes that with the rise of automation, firms’ foreign sales will further increase.

Hypothesis 4: There is a positive relationship between automation and firms’ foreign

sales.

In order to subsume the abovementioned discussion, the following conceptual model can be established (figure 4). A further explanation of the model’s control variables can be found in the study’s methodology section.

(25)

3. RESEARCH METHODOLOGY

The following section provides a description of the sample and data used as well as an explanation of the suggested research method to test the stated hypotheses. Finally, some preliminary analyses are carried out to assess the data’s quality and identify any potential problems concerning linearity, normality, homoscedasticity and outliers.

3.1. Sample and Data Collection

The empirical analysis is based on a sample of U.S. manufacturing firms derived from the ORBIS database. Firstly, the study focuses on the U.S. economy as it plays a leading as well as pioneering role in buying and leveraging new technologies (Sirkin et al., 2015). Moreover, it belongs to the five major markets of industrial automation and has experienced steady growth in robot installations since 20104 (International Federation of Robotics, 2017). Secondly, the study examines the industrial sector of manufacturing because of its substantial use of automated machines and innovative systems (Acharya et al., 2000). Analyses further show that automation’s disruptive power on the industry will increase over the next years, forcing firms to adapt their business strategies to more flexible processes (Manyika, Sinclair, Dobbs, Strube, Rassey, Mischke, Remes, Roxburgh, George, O’Halloran & Ramaswamy, 2012). In simple terms, it is the U.S. industry most affected by automation (Acemoglu & Restrepo, 2017). In addition, research assumes that industrial manufacturers operating in multiple countries are particularly vulnerable to liability of foreignness (Caves, 1982; Zaheer, 1995). Consequently, they are extremely cost-sensitive and strive to find suitable solutions to successfully meet the challenges of the future.

The manufacturing firms selected for this study belong to the North American Industry Classification Systems (NAICS 2017) food manufacturing (311), beverages and tobacco manufacturing (312), petroleum and coal manufacturing (324), chemical manufacturing (325) and computer and electronics manufacturing (334). These categories represent the five largest U.S. manufacturing industries, which are responsible for about 51 percent of manufacturing GDP (Timmons, Gold & McNelly, 2012). My study focuses on these segments because of their significant role in American manufacturing. Moreover, research found that strong industrial sectors are the most powerful drivers of innovation and economic growth (e.g. Uzkurt, Kumar,

4_{According to the International Federation of Robotics (IFR), the U.S. sale of industrial robots increased}

(26)

Kimzan & Sert, 2012). Therefore, it can be expected that they play a leading role in adopting new technological advances and developments.

By narrowing the research down to a specific industry and country, my study can gain important advantages for testing the stated hypotheses. First of all, the respective firms are very similar concerning their strategic environment and the challenges they face when internationalising and implementing automated machinery and processes. However, as they are operating in five different manufacturing segments, slight differences do exist. Consequently, the sample allows to show variance concerning the study’s field of interest by simultaneously reducing variance that might affect the suggested relationships.

The initial sample consisted of 2.772 American manufacturing firms, which were listed in the ORBIS database from 2014 to 2016. For the same period, company and industry data was retrieved from a variety of sources. First, the Datastream database was used to derive data on foreign production and foreign sales of the respective firms. Thompson Reuters’ Datastream is a powerful financial time series database providing financial and statistical data for firms and nations worldwide. Second, information on firms’ geographic dispersion and internationalisation pace was drawn from annual reports (cf. Vermeulen et al., 2002). Due to strict regulations of the U.S. Securities and Exchange Commission (SEC), listed U.S. firms are obliged to produce and publish very elaborated annual reports. Often, the so-called 10-K forms are supplemented with charts and reports addressed to firms’ shareholders, informing about main activities and achievements in a given year. Hence, these annual reports provide a rich data source for this research. Third, additional firm-specific data was gathered from Bureau van Dijk’s ORBIS database. The database combines data from different sources, such as annual reports, financial statements and merger and acquisitions. It was also used for cross-validation and compensation for missing values. Lastly, industry data about firms’ average tech spending was retrieved from Forrester’s annual survey on U.S. technology budgets. The American market research company’s expertise is grounded in annual surveys of more than 675.000 American business and technology leaders and data reported by the U.S. Department of Commerce, the U.S. Census Bureau, and the U.S. Bureau of Labour Statistics.

(27)

internationalisation activities, were excluded. As a result, the final sample consists of 199 American manufacturing firms and shows the following distribution of NAICS 2017:

Figure 5: Final sample. Distribution of NAICS 2017. Source: Own model.

3.2. Variable Measurement

In order to test the aforementioned hypotheses, time-series data at firm- and industry-level is collected. The analysis further covers a time period of three years (2014-2016), which is in line with earlier studies that emphasise the time firms require to successfully implement new business strategies (Kaplan & Norton, 2006; National Academy of Engineering, 1992).

3.2.1. Dependent Variables

When looking at literature, there appears to exist no widely accepted method for determining a firm’s level of internationalisation (Ramaswamy, Kroek & Renforth, 1996; Sullivan, 1994). Although many studies are using a single criterion, such as the number of markets served or foreign sales relative to total sales, to capture international expansion behaviour, this methodical approach remains highly questionable (Cavusgil, 1984; Hadley & Wildson, 2003; Sullivan, 1994; Welch & Luostarinen, 1988). Using such simple constructs to measure a phenomenon as complex as internationalisation fails to address behavioural dynamics such as firms’ ability to operate in foreign markets. Moreover, there is the risk of misrepresenting the concept of internationalisation and consequently distorting the results’ validity (Sullivan, 1994). Alternative perspectives suggest the use of multiple dimensions to assess the phenomenon of internationalisation more comprehensively (e.g. Chetty & Eriksson, 1999; Hitt et al., 2006; Reid, 1981; Sullivan, 1994).

N=199 21 8 7 68 95

Food Manufacturing (Code: 311)

Beverage & Tobacco Manufacturing (Code: 312) Petroleum & Coal Manufacturing (Code: 324) Chemical Manufacturing (Code: 325)

Computer & Electronics Manufacturing (Code: 334)

(28)

Therefore, this study employs four firm-based internationalisation measures, which are based on a variation of Sullivan’s (1994) and Vermeulen & Barkema’s (2002) composite measures. The first dimension, foreign production, can be represented by foreign assets expressed as a percentage of total assets. It reflects a firm’s dependence on owned foreign stocks and foreign resources (Sanders & Carpenter, 1998; Sullivan, 1994). The second dimension, internationalisation pace, includes the number of foreign subsidiaries that a firm establishes each year (Vermeulen et al., 2002). For the third dimension, geographic dispersion, the number of countries in which the firm has expanded during a given year is considered. This dimension constitutes a rough estimation of cultural and institutional variety firms experience (Lu & Beamish, 2001; Sahaym & Nam, 2013; Sanders et al., 1998). The fourth dimension, foreign sales, reflects firms’ export intensity and can be illustrated by the ratio of foreign sales to total sales.

Internationalisation data is gathered from ORBIS and the Datastream database, as well as from annual reports. These reports are manually searched for mentions of international expansions in the broadest sense, such as strategic alliances, mergers and acquisitions, the establishment of new subsidiaries and joint ventures.

3.2.2. Independent Variable

(29)

However, research shows that such approaches often miss to assess and quantify the respective development in an objective manner and hence do not lead to generalisable results (Kirk & Miller, 1986). Moreover, scholars repeatedly underline the importance of quantitative measures (Sethi & King, 1994; Simmons, 1994) as missing information about firms’ level of automation is found to hinder firms from increasing their technology investments (Fast-Berglund & Stahre, 2013; Miyatake & Kangari, 1993). Since literature on automation shows substantial lack of quantitative empirical research (Balfe et al., 2015; Bhaduri & Ray, 2004), the following study aims to bridge this gap by employing an alternative quantitative measure that includes the concept’s multidimensionality, analyses the trend of automation on the firm-level and generates results that apply to larger groups of firms. In order to capture the high complexity associated with the measurement of firms’ use of innovative technologies and processes, a composite automation variable is developed.

(30)

As shown in figure 6, my study bases its approach on three pillars which have already been used and verified to measure firms’ level of automation. By combining the following concepts, this study aims to go beyond existing research approaches and respond to current demands to intensify quantitative automation research.

Composite measures are frequently used to capture latent variables, which cannot be measured directly but rather inferred from other, observable variables (Hagedoorn & Cloodt, 2002; Schmidt & Kaplan, 1971). Since automation belongs to this group of variables, this study follows prior research and integrates several indicators which allow to assess the respective concept through a more profound and informative composite variable (Coffey & Thornely, 2006; Zhai, Goodrum, Haas & Caldas, 2009). Moreover, the study’s automation variable goes beyond research’s current focus on single industries within manufacturing, such as automotive (Coffey et al., 2006) or construction (El-Mashaleh, O’Brien & Minchin, 2006; O’Connor & Yang, 2004; Zhai et al., 2009), and aims to focus on a broader set of U.S. manufacturing firms to generate generalisable results. It further responds to scientific demands to capture automation’s multidimensionality in a holistic manner (Hervas-Oliver, Albors Garrigos & Gil-Pechuan, 2011; Nam, Parboteeah, Cullen & Johnson, 2014) instead of using single criterion measures (cf. Lu & Wang, 2018). Following Greenan (2003), both automation input- and output variables are introduced, which are found to lead to more complete and robust empirical results (Hou, Hu & Yuan, 2017). By doing so, the study does not only capture firms’ automation decisions, but also organisational responses to technological change (Azar & Ciabuschi, 2016; Chen, Podolski & Veeraraghaven, 2017). This is in line with Greenan (2003) and McDermott et al. (1999), who state that manufacturing firms invest in advanced automation technologies to achieve substantial operational, organisational and competitive benefits. Research further investigated that automation benefits are increasing as firms’ level of automation develops (Balfe et al., 2015).

(31)

factors that are frequently investigated to be connected to automation and thus describe the respective trend (cf. McDermott et al., 1999). Consequently, the automation variable is built on four components that constitute observable operational, organisational and competitive effects deriving from manufacturing firms’ increasing modernisation and automation processes and thus combined represent a good proxy for automation.

First, in order to grasp operational automation outcomes, this study focuses on labour productivity, which, in contrast to overall firm productivity (Brynjolfsson, 1993), is found to be a good measure to track firms’ improvements in production deriving from increased automation (Lannelongue, Gonzalez-Benito & Quiroz, 2017; Siegel, 1994; Zairi, 1992; Zhai et al., 2009). Labour productivity is defined as the ratio of total output to operating inputs (Samuelson & Nordhaus, 1989) and provides information about employees’ production effectiveness. As in previous research (Koch & McGrath, 1996; Lannelongue et al., 2017), this study uses the logarithmised ratio of net sales to the number of employees to measure firms’ workforce productivity.

Second, organisational automation outcomes are represented by changes in firms’ number of employees (cf. Fahrenkrog & Kyriakou, 1996; Frey & Osborne, 2016; Zairi, 1992). According to literature, firms’ increased spendings in automation lead to modified organisational structures and consequently to a decline in manufacturing employment, which is reflected in falling employee numbers (Brynjolfsson et al., 2011; Charles, Hurst & Notowidigdo, 2013; Jaimovich & Siu, 2012).

Third, this study captures competitive automation outcomes by focusing on firms’ profit margins (cf. McDermott et al., 1999; Ramamurthy, 1995). Research shows that most firms consider automation as an essential means to strengthen their competitive position and achieve sustainable competitive advantage (Acharya et al., 2017; Chung, Yam & Chen, 2004; Wei, Song & Wang, 2017). Moreover, there is empirical evidence that most investment decisions in manufacturing are based on profitability considerations (Azzone & Bertelè, 1989). This is in line with Berkeley (1985), who states that all profits are related to automation nowadays. Babbar and Rai (1990) further emphasise that automation is only successful when implemented well. This can be best assessed by looking at firms’ rising profits that are proven to indicate implementation effectiveness (Denison & Mishra, 1995; Mathur, Dangayach, Mittal & Sharma, 2011; McDermott et al., 1999; Ramamurthy, 1995).

(32)

measure of average tech spending per industry (Bartels, Guarini, Klehm & McPherson, 2016). Depending on the specific manufacturing industry firms are operating in, they experience different levels of competitive pressures to invest in advanced manufacturing technologies and make human and structural adjustments (Dereli, 2015; Mendi & Costamagna, 2016). Consequently, they differ in their automation inputs, namely technology budgets, which are expressed as a percentage of firms’ gross outputs.

Last but not least, this study focuses on micro- and meso5-level data to get closer insights about U.S. manufacturing firms’ level of automation (c.f. Kotarba, 2017; Zhai et al. 2006). This is in contrast to many composite measurements developed by neighbouring related research fields, which use macro-level data to capture complex technological constructs, such as digitization and innovation, on a regular basis (cf. Cheng & Yang, 2017; Katz & Koutroumpis, 2013). However, such approaches mainly focus on country- and social variations in particular behaviour and hence do not observe developments within specific firms and industries. While firm-level data on automation is retrieved from the ORBIS database and annual reports, data on industries’ average tech spending is collected from Forrester’s annual survey on U.S. technology budgets. Thereby, the research and advisory firm also relies on data reported by the U.S. Department of Commerce, the U.S. Census Bureau, and the U.S. Bureau of Labour Statistics. As the respective database is founded on the NAICS 2017 this study uses, too, it is possible to assign each of the study’s NAICSs a corresponding counterpart in the Forrester database (food, beverages and tobacco: consumer products; petroleum and coal: oil and gas; chemicals: chemicals; computers and electronics: high-tech products).

In order to capture firms’ level of automation, the new automation variable is computed by combining the four respective variables and calculating their mean (Field, 2009). My study further controls for the automation variable’s robustness by z-transforming and therefore standardising all variables of interest before computing the composite variable and executing all necessary analyses again (Steiger, 1980)6. As results were the same, the composite automation variable was found to be robust.

5_{When considering the economic environment, one can distinguish between three levels of analysis: micro,}

meso and macro. The meso-level consists of “market forces”, which influence firms’ business behaviour. Examples are, amongst others, specific industries, supply and demand, competitors and strategic

(33)

3.2.3. Control Variables

To accurately test the hypotheses that relate advances in automation to firms’ international expansion behaviour, it is necessary to control for a number of firm-specific figures that are found to affect internationalisation (cf. Vermeulen et al., 2002).

First, firm size is included in the analysis. According to Tushman and Romanelli (1985), large firms can benefit from their personnel’s specialised and comprehensive knowledge when expanding into foreign markets. Furthermore, they are associated with a strong ability and wealth of resources to deal with complex business challenges (Franko, 1989; Henderson & Fredrickson, 1996). Consequently, it can be assumed that it is easier for larger firms to internationalise. Firm size is measured by firm’s total assets in a given year.

Second, firm age is incorporated by measuring the number of years since the firm’s foundation. Research shows that older firms, in contrast to younger ones, are better embedded in transnational business networks and have larger internationalisation market commitments needed for international expansion. In addition, they possess better developed organisational resources, which enable quicker adaption to foreign markets (Yip, Biscarri, & Monti, 2000). Thus, firm age may impact internationalisation decisions.

Finally, the study controls for firms’ financial structure by using the debt ratio of total liabilities to total assets (Hitt et al., 1997; Vermeulen et al., 2002). In reference to Jensen (1986), firms’ ability to expand into foreign markets is highly dependent on their capital structure and hence on their financials.

(34)

Table 2 presents a summary of all variables, measures and data sources this study uses.

Table 2: Variables, measures and sources of data

Variable Measure Source

Dependent Variables Foreign production Ratio foreign assets to total assets ORBIS/ Datastream

Internationalisation pace Number of internationalisation activities a firm performs in a given year

Annual reports

Geographic dispersion Number of countries in which the

firm expands during a given year Annual reports Foreign sales Ratio of foreign sales to total sales ORBIS/ Datastream

Independent Variable Automation Composite variable, consisting of:

§ Labour productivity § Changes in firm’s nr. of

employees § Profit margin

§ Average tech spending per industry

ORBIS Annual reports Forrester database

Control Variables Firm size Firm’s total assets ORBIS/ Datastream

Firm age Number of years since the firm’s

foundation ORBIS/ Datastream

Financial structure Ratio of total liabilities to total assets ORBIS/ Datastream

3.3. Research Model

When analysing the empirical relationship between two continuous variables, most studies use the principle of bivariate correlation, notably Pearson’s correlation coefficient (Ezekiel, 1930). Despite the meaningful information derived from the examination of the respective correlation coefficient, the measure lacks the capacity to provide information about whether additional factors eventually influence the observed relationship between two variables. To overcome this issue, my study introduces partial correlation (Baba, Shibata & Sibuya, 2004). This statistical concept measures the degree of correlation between two variables while simultaneously removing the effects of one or more related control variables and thus avoiding misleading information (Field, 2009). In this way, spurious correlation between two variables can be excluded.

(35)

process is introduced. In reference to Faller (1981), this study uses Fisher’s weighted mean correlation coefficients to calculate the average correlation coefficients of the respective time series. One prerequisite for this methodical approach is to compute partial correlations between each pair of time courses. Subsequently, as correlation coefficients are not additive (Kendall & Stuart, 1979; Silver & Dunlap, 1987), they are transformed into new variables called Fisher Zs. In order to do so, Fisher’s Z transformation (Fisher, 1921), which consists of an elementary transcendental function called the inverse hyperbolic tangent function, is used:

Formula 1:

! =1 2×&'

1 + ) 1 − )

Fisher’s Z transformation almost entirely corrects the skew in the distribution of r. Consequently, the newly generated Fisher Z variables are approximately normally distributed and can be added for calculating a mean value. After the average Z scores ! are calculated, the Fisher weighted mean correlation coefficient ) is created by backtransforming the average Z variables to the form of r:

Formula 2:

) = +,×-− 1 +,×-_{+ 1}

In reference to Silver et al. (1987), who investigated different procedures for averaging correlations via Monte Carlo simulations, Fisher’s Z transformation is the methodical instrument most precise and least prone to bias. Moreover, it can be applied for all sample sizes.

3.4. Preliminary Analysis

(36)

To assess linearity between the relevant variables, scatter plots are created. Thereby, the dependent variable is placed on the x axis whereas the independent variable is represented on the y axis (Field, 2009). Almost all scatter plots created indicate weak and lightly observable linearity, except foreign sales7. This observation might be an interesting indicator for the subsequent main analysis, which focuses on the respective relationships in more detail.

In order to detect normality, the Shapiro Wilk test is applied because of its high statistical power (Field, 2009). Besides testing the null hypothesis that the respective sample comes from a normally distributed population, this study further measures data’s kurtosis and skewness (Ghasemi & Zahediasl, 2012). According to Field (2009), perfect normal distribution is given if values of both skewness and kurtosis equal zero. However, current research shows that skewness and kurtosis between +/- 1.96 are considered acceptable in order to prove normality, too (George & Mallery, 2010; Gravetter & Wallnau, 2014). Moreover, normal Q-Q (quantile-quantile) plots are created. Thereby, the variables’ probability distributions are compared by plotting their quantiles against each other (Fowlkes, 1987).

When looking at the Shapiro Wilk test’s results, all p values are < .05 and hence no variable can be identified as normally distributed. However, as shown in table 3, further analyses relating to skewness and kurtosis demonstrate a slightly different picture. In reference to George et al. (2010), all foreign sales variables seem to be normally distributed as their skewness and kurtosis values are within the acceptable range of +/- 1.96. The same applies to the years of existence variable, which fulfills the normality-assumption for all three years. Moreover, there are two variables which show mixed results: Whereas all automation variables show acceptable skewness over the time period observed, they have higher positive kurtosis levels than allowed. In addition, geographic dispersion variables for the years 2014 and 2015 present acceptable skewness values while simultaneously exceeding the acceptable range for kurtosis. All other variables show skewness and kurtosis figures above or below the acceptable range of +/- 1.96. The graphic visualisation of Q-Q plots8_{confirms the analyses’ premature}

results and shows that both foreign sales and years of existence variables are normally distributed. Moreover, the diagrams strengthen automation’s normality assumption for all years observed while simultaneously showing that there exist some outliers. Regarding geographic dispersion, the Q-Q plots demonstrate a slightly normal distribution for the time period 2014-2016. For the remaining variables, no normal distribution is found.

(37)

When considering the results, it can be inferred that a large number of variables are not normally distributed and, hence, the assumption of normality has been violated. However, referring to Ghasemi et al. (2012) and Salkind (2010), normality is only a consideration when the sample size is very small. Consequently, as the study’s sample is > 40 (N=199), the normality assumption is not needed. According to Ghasemi et al. (2012), it is the Central Limit Theorem which ensures that the disturbance term’s distribution will be close to normality in sufficiently large samples.

Table 3: Normality Test

(38)

Homoscedasticity describes a situation in which the variance of residuals is the same across all values of the independent variables. It can be tested by creating and analysing scatterplots of the standardised residuals (y-axis) against the predicted values of the dependent variable (x-axis) (Kutner, Nachtsheim & Neter, 2003). Although some outliers can be detected for the respective variables, the overall observation confirms homoscedasticity for all variables.9 Consequently, the homoscedasticity assumption can be accepted.

Finally, in order to identify potential outliers, the study uses the graphic visualisation of box plot diagrams. Thereby, each data point, which deviates more than 1.5 standard deviations from the sample’s mean, constitutes a light outlier. Extreme outliers are marked by starlets and deviate more than three standard deviations. When looking at the created box plot diagrams, all variables, except foreign sales, show a number of outliers. While foreign assets, geographic dispersion and internationalisation pace possess outliers being higher than the sample’s mean for all years observed, outliers concerning automation move in both directions.10 Extreme outliers are mostly found for the foreign assets variable, but also automation, geographic dispersion and internationalisation pace show occasional outliers deviating more than three standard deviations from the sample’s mean. However, as all observations that were labelled as outliers were manually checked and verified, the study concludes that none constitute an error. This is in line with prior research, which states that a small number of outliers is quite usual in large samples and does not result from any anomalous condition (Aguinis, Gottfredson & Joo, 2013). Salkind (2010) further states that valid outliers can be retained as they reflect a realistic picture of the targeted population. However, to make sure that the detected outliers do not distort the research’s results, the study conducts its main analysis with variables of which the outliers were removed once again. As the respective approach produced similar results, but reduced the model’s explanatory power11, this study decided to retain all outliers.

9_{See Figure I.13.}

(39)

4. RESULTS

4.1. Descriptive Statistics

When looking at the study’s final sample, it can be observed that it comprises a wide range of American manufacturing firms. Firstly, the average firm size, represented by a firm’s total assets, is $16,030,195.87k. As shown in table 4, which refers to the entire time period observed,12 the largest firm in terms of assets is valued $349,493,000.00k whereas the smallest firm owns $189,713.00k. This large span is also represented in the variable’ standard deviation, which displays $38,824,857.00k. Secondly, concerning firm age, the average firm is found to be 54.12 years old. However, as the variable’s standard deviation is relatively high (46.18), it can be assumed that the sample contains a good mixture of both young and old firms. According to analysis, the oldest firm was incorporated 214 years ago. The minimum value is 0, which indicates that some firms just started their business in the study’s first year of observation. However, further analysis showed that some of the respective firms emerged from already existing organisations and weren’t therefore completely new to the market. Thirdly, in order to assess firms’ financial structure during the defined time period, my study looks at firms’ debt ratios. This variable has a mean of 0.56 and a standard deviation of 0.28. Consequently, it can be said that the average firm has a debt level of 56%, which reflects the high levels of operating leverage manufacturing firms face. The minimum value observed is 0.03 whereas the maximum value is 2.81. Thus, the sample consist of both firms with very low debt and a few with debt exceeding their total assets.

Table 4: Descriptive Statistics for the time period observed

Variables Minimum Maximum Mean Std. Deviation

Total Assets 189,713.00 349,493,000.00 16,030,195.87 38,824,857.00 Firm Age 0 214 54.12 46.18 Financial Structure 0.03 2.81 0.56 0.28 Foreign Assets 0.00 105.32 15.32 20.04 Internationalisation Pace 0 27 2.13 3.20 Geographic Dispersion 0 14 1.44 1.76 Foreign Sales 0.00 100.10 53.03 24.10 Automation -13.10 20.52 7.66 2.99

Science fiction as a business fact.