Time, geography and science : how to manage non-local search for innovation

(1)

Time, Geography and Science:

How to Manage Non-Local Search for Innovation. Final Master’s Thesis

Student/ Student’s No: Esther Plagmeijer / 10615873

University of Amsterdam, Amsterdam Business School Qualification: MSc. in Business Studies – Strategy Track

Supervisor: dr. R.M. (Ranjita) Singh

University of Amsterdam, Faculty of Economics & Business – Section International Strategy & Marketing

(2)

Abstract

Innovation has long been recognized as the main source of productivity growth and industry development. In order to innovate, firms tend to generate and acquire new knowledge. The search for new knowledge happens along two distinct paths: via local search and non-local search. The concept of search expands the existing literature regarding the tension of exploitation and exploration. Several research studies advocate for the adoption of both local and non-local search, though few have actually pursued in-depth research regarding the specific concept of non-local search. Authors have been rather inconclusive on the types of non-local search and the effect that they have on the innovative performance of the firm. In an attempt to advise firms where to invest, this Master’s thesis elaborates on the concept of non-local search in terms of time, geography and science. Following the patent activity of the transport wheelchair industry, this research finds that the impact on the innovative capacity of the firm is highest during non-local search for science. Non-local search for science refers to the technological diversification undertaken by the firm. The impact seems to hold true for different industry periods, both turbulent, and calm and consistent. The findings prove to be beneficial to both theory and practice. In particular, the results provide a framework for managers to utilize in their non-local search attempts.

(3)

Introduction

Albert Einstein: ‘Invention is not the product of logical thought, even though the final product is tied to a logical structure.’ (Pais, 1982, p. 131)

‘A new competitive landscape is developing based on the technological revolution and increasing globalization’ (Hitt, Keats, & DeMarie, 1998, p. 22). Faced with complexity, firms struggle to create and maintain competitive advantage, especially within industries with a high dependence upon technology. In order to advance, compete and deliver, firms rely on their ability to innovate. Innovation includes the ‘entire process in which firms transform ideas into valuable and scarce products, services or processes’, and requires a combination of existing and new knowledge (Galbreath, 2005) (Rowley, Baregheh, & Sambrook, 2009, p. 1334) (Howells, 2002). Search for new knowledge happens along two dimensions; via local and non-local search. Whereas local search centers on path deepening and is likely to be constrained to the current, existing areas of search, non-local search focuses more on path widening. Non-local search moves beyond local search by operating along new, non-existing paths (Rosenkopf & Nerkar, 2001). Researchers (Katila & Ahuja, 2002) generally recommend firms to adopt both local and non-local search in order to stimulate balanced, but unique innovations. Despite existing literature, the particular concept of non-local search remains vague and inconclusive. What types of non-local search exist? How do they relate to the innovative performance of the firm? These questions guide the innovative behavior of the firm, yet they remain unanswered. Additionally, research specifically calls for the introduction of longitudinal studies operating in complex markets (Rosenkopf & Nerkar, 2001). Eventually, such studies will help to discover what actually determines firm success in a particular industry. This research area seems to be explored by many, yet it is still not completely understood.

(5)

This master thesis provides an empirical study with respect to non-local search. The study uses both filed and granted patent data from the US Patent and Trademark Office, specifically from the Transport Wheelchair Industry. The study performs a quantitative analysis of the types of non-local search and their individual contributions to firm innovation. Central to the study are three types of non-local search: time, geography and science (Poldolny & Stuart, 1995) (Stuart & Poldolny, 1995) (Chuang & Baum, 2003).

Overall, this study aims to contribute to the proposed gap in the literature of innovation and search. Second, this study aims to advise firms where to invest, in order to create favorable advantages within the industry. The research question, therefore, is:

‘How does non-local search help a firm to innovate within a particular industry?’

The structure of the master thesis is as follows; first, a critical review of the existing literature is given, after which the research gap and research question are addressed. Subsequently, the conceptual framework introduces the hypotheses and underlying constructs. This is followed by a thorough discussion of the research method and results. Finally, the thesis ends with a discussion, which will provide managerial implications, limitations and areas for further research.

(6)

Literature review

This section presents the most substantial insights found in the literature of innovation and search. First, the concept of innovation in the literature will be discussed. In particular, the literature will show a link between innovation and industry evolution. Subsequently, this section will pay attention to the concepts of organizational knowledge and search. Finally, the resulting research gap discovered in the literature will be considered, along with its impact on the central research question of this Master’s thesis.

Innovation and the Evolution of industry

Much of the current literature seems to agree with Grant (1991, p.114) and defines organizational strategy as the ‘match between internal resources and skills, and external risks and opportunities, performed by the firm’. In this view, various theories of the firm try to capture the reason why firms design strategies as they do (Grant R. , 1991) (Grant R. , 1996). The theories have come a long way: starting with a neo-classical approach, in which firms exist as input combiners, and moving towards a more resource-centered approach, in which firms exist in order to seek expensive-to-copy inputs (Conner, 1991). Along with these developments, competitive advantage has made its appearance. Due to the heterogeneity and immobility of resources, firms are now better able to create competitive advantage within the marketplace (Barney, 1991) (Conner, 1991) (Grant R. , 1996). In order to advance, compete and differentiate, firms tend to innovate. Innovation resembles ‘the multi-stage process in which organizations transform ideas into new/improved products, services or processes’ (Rowley, Baregheh, & Sambrook, 2009, p. 1334). Innovation tends to reflect the entrepreneurial spirit of the firm: it reflects the engagement in product-market innovation, the operation of risky ventures and the discovery of pro-active innovations (Thornhill, 2006). Different types of innovation do exist, such as incremental innovation, modular innovation,

(7)

architectural innovation or radical innovation (Henderson & Clark, 1990). The work of Joseph Schumpeter probably centers most on the concept of innovation and performance. His notion of creative destruction resembles ‘the process by which innovations tend to replace older ones’ (Aghion, Akcigit, & Howitt, 2013, p. 1). Schumpeter notes how innovation can be a powerful tool for firms as they try to enter the market with their emerging or disruptive technologies (Cefis & Marsili, 2005). This behavior tends to undermine the success of established firms who, therefore, are also required to innovate. From this, it seems that innovation is a powerful tool, mostly to create productivity growth and industrial development (Abernathy & Clark, 1985).

Despite following different paths of innovative development, firms and industries do pursue several phases that can be identified as common to many (Strebel, 1987). Strebel (1987) notes how a typical industry tends to evolve from a rather radical, discontinuous innovation, as the result of fierce and independent entrepreneurial effort. Clark (1985, p.236) refers to this period as the ‘fluid phase; performance criteria are not yet defined, and market needs and process difficulties are approached through a variety of product designs’. Innovation, by itself, is observed as being relatively rapid and fundamental. After this initial phase, a particular design is expected to achieve dominance in the market, and process and performance criteria tend to be better specified (Clark, 1985). This development calls for the introduction of dominant design. Dominant design resembles ‘a specific path, along an industry’s design hierarchy, which has successfully established dominance among competing design paths’ (Suarez & Utterback, 1995, p. 416). The firm that achieves dominant design has managed to develop favorable control over its market and technological factors (Peteraf, 1993). Reaching dominant design is beneficial because it creates a positive effect on long-term market share, profitability and, ultimately, firm survival (Suarez & Utterback, 1995).

(8)

This transition towards dominant design calls for the ‘specific phase’ (Clark, 1985, p. 236). In this second phase, gains are to be made by the application of process efficiency, the use of complementary assets or the possession of a strong regime to appropriate profit (Clark, 1985) (Utterback & Abernathy, 1975). Complementary assets include both services that can commercialize the asset as well as the products and services that help maximize the value of the asset (Linden, Kraemer, & Dedrick, 2009) (Teece, 1986). The appropriability regime refers to ‘the environmental factors that govern the innovators’ ability to capture profits generated by the innovation’ (Linden, Kraemer, & Dedrick, 2009) (Teece, 1986, p. 287). Examples of such mechanisms are patents, copyrights and trade secrets. Teece (1986) also refers to these two stages of evolutionary development. He labels them as ‘the pre-paradigmatic and pre-paradigmatic phases’ (Teece, 1986, p. 287). Competition, while boosted in the first phase, is eventually slowed down in the second phase. This is mainly due to the introduction of dominant design and the lack of competition for market share. Strebel (1987) notes that when the incremental stream of products and processes stops, growth will eventually start to decline.

Established industries often find themselves in the second phase of evolutionary development, the specific or paradigmatic phase. Dominant design has been established and firms compete via incremental innovation or by shifting to a new innovation curve. In the literature, much attention is paid to the relationship between innovation and firm performance during the evolution of a particular industry. Achievements in the industry, due to dominant design or incremental innovations, seem to be linked to the firm’s ability to innovate and to capture value from this innovation. But how do firms realize these achievements? Knowledge on this topic seems to be ever fruitful, mainly to determine what actually accounts for a firm’s performance in a particular industry. This concept, though widely studied, remains incompletely answered.

(9)

Innovation, Knowledge and Non-Local search

Firms tend to create competitive advantage by maintaining valuable, rare and in-imitable resources (Barney, 1991) (Hall R. , The strategic analysis of intangible resources, 1992). The actual source of this advantage stems from the intangible resources possessed by the firm (Hall R. , The strategic analysis of intangible resources, 1992). Intangible resources can be characterized as either assets or competencies (Hall R. , 1993, p. 608). Assets include all types of intellectual property rights, such as patents, trademarks and trade secrets. Competencies include the collective attributes that add up to the organizational knowledge and expertise of all internal stakeholders (Hall R. , 1993, pp. 608- 609). The main determinant of intangible resources is knowledge (DeCarolis & Deeds, 1999). Knowledge represents ‘the dynamic framework in which information can be stored, processed and understood’ (Howells, 2002, p. 872). Knowledge can be categorized by knowledge stock or knowledge flow. ‘Knowledge stock denotes the collection of individual knowledge and competencies at a single point in time, while knowledge flow reflects the repertoire of declarative knowledge and procedural knowledge’ (Thornhill, 2006, pp. 691- 692). Innovation calls for the adoption of both these types of knowledge. Whether a firm is actually able to capture value from its own knowledge base is primarily determined by the actual transferability of knowledge, the capacity for its aggregation and the appropriability of the knowledge-holder (Grant R. , 1996). In order to innovate, invent and discover, firms use existing knowledge or generate and acquire new knowledge (Howells, 2002). The creation of new knowledge happens either via acquisition or via the combination of existing knowledge. The search for new knowledge occurs by way of local and non-local search. Local search relates to the concept of exploitation and it is characterized by refinement, efficiency and selection. Non-local search relates to the concept of exploration, and it is characterized by search, variation, risk and

(10)

implementing exploitation or exploration (March J. , 1991). Maintaining the right balance between the two tends to be difficult due to the different levels on which the constructs operate; for example, the individual, organizational or social system level (March J. , 1991).

Local search is specified as being path deepening, and it is likely to be constrained to the neighborhood of current, existing search (Ahuja & Katila, 2004, p. 888). Firms that apply local search tend to introduce research and development (R&D) activity that is similar to their previous activities (Rosenkopf & Nerkar, 2001). Stuart and Podolny (1996) indicate how a firm engages in search when ‘its niche shifts across time periods, and the manifestation of ‘localness’ is equivalent to the amount of the shift in its niche’ (Stuart & Podolny, 1996, p. 21). The firm that applies local search finds itself in markets where only few firms dominate (Martin & Mitchell, 1998). Local search markets are characterized by conditions of high opportunity and appropriability. This allows incumbents to accumulate technological knowledge (using Schumpeter Mark II technologies) and innovative capabilities, in order to innovate and build advantages over new entrants (Malerba & Orsenigo, 1996, p. 452). Katila & Ahuja (2002, p. 1184) indicate how local search can positively affect firm innovation in three ways. First, similar knowledge reduces the likelihood of errors and false starts and it facilitates the development of routines, making the search more reliable. Second, as the knowledge seems familiar, the search is developed as being predictable. And finally, a deeper understanding of the concept will boost the firm’s ability to identify valuable elements (Katila & Ahuja, 2002, p. 1184). Katila (2002) adds that old extra-industry knowledge actually stimulates the knowledge base of the firm. Though, effective in the short run, local search can be rather self-destructive in the long run (March J. , 1991). Rosenkopf & Nerkar (2001, p. 289) show that a focus on first-order competence – the distinctive competence of the firm – can lead to the development of core rigidities or competency traps. Eventually the firm will face convergence of its search approaches and will face a decline in innovation due to

(11)

diminishing returns and lack of search space (Ahuja & Katila, 2004, p. 888) (Katila & Ahuja, 2002). New search space must therefore be explored in order to improve the technological trajectory of the firm (Ahuja & Katila, 2004).

Non-local search is specified as being path widening or path creating, and it moves beyond local search opportunities to reconfigure the current knowledge base (Rosenkopf & Nerkar, 2001). This type of search allows search along new, non-existing paths and is the fundamental mechanism by which firms gain new knowledge (Rosenkopf & Nerkar, 2001). Firms tend to apply non-local search in mature markets, where they have to overcome the loss of market share to other firms (Martin & Mitchell, 1998). Many firms tend to be active in these markets; the innovation base is continuously enlarged through the entrance of new firms and dissolution of established firms (Malerba & Orsenigo, 1996). Non-local search markets are characterized by conditions of high opportunity and low appropriability. These conditions favor a continuous supply of new entrants, blocking enduring innovative success for the firm (Malerba & Orsenigo, 1996, p. 452). By increasing the scope in R&D, non-local search tends to result in an increase in total firm knowledge. Especially while adding new knowledge elements to the current knowledge base also results in an increase in new combinations and recombinations (Rosenkopf & Nerkar, 2001). Despite possessing the ability to reach far greater returns than with local search, non-local search seems systematically less certain, more isolated in time and more distant from the firm’s core business (March J. , 1991).

Innovation researchers have dedicated much attention to the topic of search. While early contributions focus on the tension between exploitation and exploration, more recent additions emphasize the concepts of innovation and search. Search studies generally advise the adoption of both local and non-local search in order to provide the right balance in R&D. Katila and Ahuja (2002, p. 1186) reiterate this conclusion by suggesting to combine

(12)

firm-order to create original, unique combinations. Focusing too much on either local or non-local search is viewed as inefficient. Katila and Ahuja (2004, p. 887- 890) identify two circumstances that can create new search paths: geography and science. They indicate the presence of a curvilinear (inverted U) relationship between non-local search in geography and science and the subsequent innovativeness of the firm. Tushman and Anderson (1986) conclude that non-local search is often initiated by new firms and that it creates environmental turbulence within the industry. The new firm that initiates the changes tends to grow more rapidly than other existing firms. Additionally, Rosenkopf and Nerkar (2001, p. 287) indicate that the impact of exploration on technological development is greatest when exploration spans both organizational and technological boundaries.

Previously, it was determined that knowledge provides the actual feedstock for innovation. The actual search for this knowledge is therefore very important to the firm. Despite the existing literature, the concept of non-local search remains vague and inconclusive. Research has determined that non-local search, under some circumstances, has the ability to strengthen the innovative capacity of the firm. This, while non-local search tends to enlarge the current knowledge base with new, non-existing knowledge. But what types of non-local search exist? And how do these types relate to the innovative performance of the firm? These questions remain unanswered. In particular, research indicates a need for longitudinal studies, operating in complex markets, where innovations seem to be the key determinant of success (Katila & Ahuja, 2002) (Rosenkopf & Nerkar, 2001). Rosenkopf and Nerkar (2001) advise researchers to compare and contrast firm behavior in different kinds of technological contexts, mainly to generalize results.

(13)

Literature gap and Research question

The above literature review suggests a gap in the literature on innovation and search. Specifically, literature has limited empirical research regarding the types of non-local search and the effect that they have on the innovative capacity of the firm. This has resulted in an insufficient managerial knowledge of where to invest in order to maintain a high level of innovation. This study aims to contribute to this gap in the literature by researching non-local search in terms of time, geography and science.

While contributing to the literature, the proposed research question therefore is:

(14)

Conceptual Framework

This section identifies the concepts that underlie this Master’s thesis. Based on the literature and research objectives, a total of four hypotheses are presented.

Hypotheses

The conceptual model displayed in Figure 1 at the end of this section shows the four hypotheses for this Master’s thesis. As the figure demonstrates, non-local search is defined in terms of time, geography and science, and acts as an independent variable. The dependent variable is labeled firm innovation, and is defined by the number of forward references and the range of industries in which the patent operates. Existing literature has already determined a link between non-local search and performance (Ahuja & Katila, 2004) (Katila & Ahuja, 2002). However, the effect of the individual dimensions of non-local search on innovation seems to be rather unexplored. In order to contribute to this proposed gap, four hypotheses are discussed below. It is important to mention that this Master’ thesis does not attempt to compete with existing search studies in relation to the effect of non-local search on innovation. Rather, this thesis accepts existing findings (the curvilinear effect between non-local search and innovation) and aims to go more in-depth regarding the types of non-non-local search and the effect that they have on firm innovation.

Non-local search in Time

H1: The greater the degree of search in time, the greater the related innovative performance.

(15)

The first hypothesis mainly introduces two concepts: the concept of non-local search in time, and the concept of firm innovation. Non-local search in time implies ‘the creation of knowledge via a path-dependent evolutionary process that involves the recombination of knowledge acquired over time’ (Nerkar, 2003, p. 211). Firm innovation corresponds to the extent of innovation displayed by the firm (in terms of patents). Firm innovation reflects the ‘generation, acceptance, and implementation of new ideas, processes, products and services’ (Calantone, Cavusgil, & Zhao, 2002, p. 15). Firms are entitled to innovate in order to cope with their aggressive competitive environments.

While exploring the relationship between non-local search in time and firm innovation, the concept of organizational learning arises. According to organizational learning, firms learn by ‘encoding inferences from history into routines that guide behavior’ (Levitt & March, 1988, p. 319). New knowledge, then, would reflect the outcomes of old knowledge, and acts in some sort of sequential order (Sanger & Levin, 1992). Research (Calantone, Cavusgil, & Zhao, 2002, p. 516) has supported this relationship; organizational learning relates to the development of new knowledge, something that is crucial for the innovative capability and performance of the firm. In his 2003 study, Nerkar (2003, p. 212) identified three approaches related to the evolution of knowledge inside the firm; ‘the rational non path-dependent approach, the random path-dependent approach and the bounded rational path-dependent approach’. The second approach relates to this hypothesis; with the random path-dependent approach, knowledge seems to be the actual outcome that emerges from a rational process operated by the firm (Nerkar, 2003, p. 212). Despite the fact that old knowledge supports innovation, mainly because of its reliability and legitimacy, some researchers remain skeptical (Katila R. , 2002). According to them, firms must build on the most recent knowledge, mainly because old knowledge would make innovation absolute

(16)

This first hypothesis predicts that non-local search in time has a positive effect on the innovative performance of the firm. The firm that actively uses old knowledge, and thus is committed to organizational learning, is likely to possess state-of-the-art technology, leading to high innovative performance (Calantone, Cavusgil, & Zhao, 2002).

Non-local search in Geography

H2: The greater the degree of geographic search, the greater the related innovative performance.

The second hypothesis introduces, again, two concepts: the concept of non-local search in geography, and the concept of firm innovation. Non-local search in geography implies the search for innovation across geographic borders, in unknown areas. Firms that perform geographic search are mainly driven by local opportunities and problems (Ahuja & Katila, 2004). Again, firm innovation reflects the extent of innovation exhibited by the firm and relates to the ‘generation, acceptance, and implementation of new ideas, processes, products and services’ (Calantone, Cavusgil, & Zhao, 2002, p. 15).

Firms tend to be limited contextually in their search for new knowledge due to experience and expertise (Almeida & Rosenkopf, 2003). Distant contexts, outside the organizational and relational boundaries, are able to provide new insights to the firm (Almeida & Rosenkopf, 2003). Ahuja and Katila (2004) expand upon this positive effect of geographic search on firm knowledge and innovation via two dimensions. First, international presence tends to raise awareness regarding the different areas of the knowledge landscape, and with that, provides a broad range of material for knowledge (Ahuja & Katila, 2004, p. 892). Adding raw knowledge can result in new, valuable knowledge combinations for the

(17)

firm. Second, international presence helps the firm to link to the existing regional networks of knowledge in a faster way than the market mechanism does (Ahuja & Katila, 2004, p. 892). Asheim and Isaksen (2002, p. 77) expound upon this same regional network. According to their study, the collaborative regional network has decisive significance for the innovative activity of the firm. In order to strengthen their competitiveness, firms tend to balance local knowledge with the existing external knowledge (Asheim & Isaksen, 2002, p. 85). In order to tap into these local knowledge networks, firms tend to mobilize inventors or form strategic alliances (Almeida & Rosenkopf, 2003). Ahuja & Katila (2004) propose a curvilinear effect between local search in geography and innovation. Firms are entitled, by applying non-local search in geography, to operate in different markets that tend to possess different user needs. Searching across organizational and relational boundaries can therefore result in much inefficiency. Second, searching across borders can lead to difficulties with respect to integration and centralization (Ahuja & Katila, 2004).

This second hypothesis predicts that the geographic search of the firm is positively related to its innovative performance. The firm that taps into local regional networks, across borders, will retrieve raw material for new knowledge, and combinations. This eventually leads to high innovative performance.

Non-local search in Science

H3: The greater the degree of search in science, the greater the related innovative performance.

(18)

the extent to which the innovative activity of the firm spans more than one technology (Breschi, Lissoni, & Malerba, 2003) (Garcia- Vega, 2006). Again, firm innovation reflects the extent of innovation operated by the firm.

In order to develop new products and services, firms tend to apply a wide variety of complementary technologies. These complementary technologies intend to maintain differences in order to be most effective. Despite differences, complementary technologies are often knowledge-related; firms tend to cluster around groups of technologies that share a common knowledge base (Breschi, Lissoni, & Malerba, 2003). Garcia- Vega (2006) specifies how technology-diversified firms (companies operating these complementary technologies) tend to possess certain advantages over specialized firms. First, technology-diversified firms are able to obtain high cross-fertilization between different, but related, technologies (Garcia- Vega, 2006, pp. 230 - 231). Second, diversification can hinder a negative lock-in effect in one specific industry, and it can maintain the development of the firm (Garcia- Vega, 2006, pp. 230 - 231). Miller (2006) refers to the same advantage. He finds a ‘positive relationship between diversification based on technological diversity and market- based measures of performance’ (Miller D. , 2006, p. 601). Despite possessing advantages, some researchers (Ireland & Hitt, 1994) (Miller D. , 2006) indicate that technological diversification has a curvilinear effect on firm performance: relying too much on technological diversification prevents integration and responsiveness. Additionally, managing technology-diversified firms tends to be rather difficult due to complexity and information asymmetry. With her empirical study, Garcia- Vega (2006, p. 230) concludes that R&D intensity as well as patent activity increases with the degree of technological diversification of the firm. She reasons that the firm can receive more spillovers from other technological fields and can reduce the risk technological investments by using technological diversification. Even more important, Quintana-Garcia & Benavides-Velasco (2008, p. 492) find that technological diversification

(19)

has an even stronger effect on exploratory than on exploitative innovative capability. Exploratory innovative capability directly relates to the non-local search of the firm, and therefore is most important for this research study - -Velasco, 2008).

This third hypothesis predicts that the scientific search of the firm, in terms of technological diversification, is positively related to its innovative performance. The firm that applies technological diversification is expected to maintain high exploratory innovative capabilities.

H4: The degree of scientific search best explains the related innovative performance of the firm.

Hypothesis 4 relates to the same concepts and underlying constructs as indicated for Hypothesis 3; scientific search and firm innovation. It is proposed that, of all three independent variables – time, geography and science – the latter will have the greatest effect on the innovative performance of the firm. Companies that tend to apply a variety of complementary technologies are expected to maintain a high level of innovation. Despite the importance of committing to organizational learning via the usage of old knowledge and of retrieving raw material for new knowledge via local regional networks, technological diversity is crucial for the innovative capacity of the firm.

(20)

(21)

Methodology

This section presents the research methodology used in this Master’s thesis. First, determinants of the research design are discussed, for example, type of design and data. Second, based on the selection of sources and capitation of data, the sample design is presented. The variables are introduced, after which conclusions are made regarding the types of measurement. Finally, the importance of generalizability, reliability and objectivity is explained.

Research design

This Master’s thesis aims to explore how successful firms search for innovation, in order to compete in established markets or to enter new ones. Central to the thesis is the assumption that search occurs via two separate paths, determined as local and non-local. This study aims attention at non-local search and introduces three underlying factors: time, geography and science. Subsequently, the study explores the individual effect between non-local search in time, geography and science and the level of innovation of the firm. Researching in this way requires a mixed method research design; ‘an intellectual and practical synthesis based on qualitative and quantitative research’ (Johnson, Onwuegbuzie, & Turner, 2007, p. 113). While the relationship-testing requires a quantitative approach, the interpretation of underlying constructs and context requires a more qualitative approach. The choice of mixed method research design is expected to enhance the value of the overall study. This Master’s thesis is presented as being explanatory; it draws attention to studying a particular situation in order to explain several relationships. In order to clarify these relationships, this research chooses to make use of longitudinal/panel study. Panel study

(22)

The proposed hypotheses will be tested via the usage of a special type of experiment: the within-subject experimental design. Within-subject design is a type of quantitative experiment in which each participant (in this case, each operating patent) is subjected to every condition or treatment, in order to study the effect on the dependent variable (Hall R. , Within-Subjects Designs, 1998). The advantage of such a design is that it tends to reduce the error variance associated with individual differences, something that is important when using firm patents. It makes it easy to detect differences while every subject’s behavior under one condition is compared to another. This approach allows testing of all patents concerning their application of non-local search in time, geography and science, mainly with respect to firm innovation.

Additionally, changes in the evolution of the industry will be taken into account, via the testing of two different time frames. Despite of the limited control over extraneous variables, the natural setting of the research does contribute to high external validity. Finally, the experimental study requires a deductive research approach; testing existing theory by means of observation. The choice for mixed method design often results in the use of both deduction and induction. Induction – theory building – is proposed in relation to the findings.

Sample and data

This study will be executed with the help of the United States Patent and Trademark Office (USPTO) database. The USPTO is part of the U.S. Department of Commerce, and is responsible for the examination of the application, publication, recording and maintaining of U.S. patents. In their study, Stuart and Podolny (1996, p. 35) show how patents and patent citations represent the technological network of the industry, and thus are of importance for this study. The decision to use raw secondary data rests upon the assumption that no processing of the data has taken place. This study particularly focuses on the filed and granted patents within the Transport Wheelchair Industry (TWI). The choice of the TWI has been

(23)

carefully made based on three distinct objectives. First, the aging of the global population and the upgrading of the U.S. social system are expected to contribute interesting dynamics to this research. These two trends will most likely re-boost the TWI, due to a sudden rise in demand. Second, technological development has positively influenced the TWI. This trend promotes waves for new innovation, mostly due to an increase in new, innovative patents. Finally, the industry in itself seems rather interesting. The industry has characteristics of high modularization – components and subsystems within an assembled product interact with each other, and each firm has selected its own S-curve (Christensen, 2000). Because of this modularization, the industry can be categorized as complex. With that characterization, the choice of industry contributes to the gap identified in the literature: there is a need for longitudinal studies in complex industries where innovation is key (Rosenkopf & Nerkar, 2001).

It is important to mention that this Master’s thesis, despite of the four studies executed, chooses to aim attention at one particular research study. This study includes all granted patents of the USPTO database that operate the second dependent variable DV2RangeIndus (label: Dependent Variable 2 Range of Industries). Reason for this approach stems from the more interesting findings for this research study. Testing both the filed and granted patents strengthens the quality and validity of this research (Trajtenberg, 1990). The findings of the three other studies, one for the granted patents and two for the filed patents, can be found in Appendix G.

The first sample of patents drawn from the database will encompass a time frame of 13 years, ranging from 1990 to 2003. This specific time frame reflects the above trends. Due to measurement difficulties, the limit for the time frame is 2003; forward and backward citations include plus and minus ten years.

(24)

Variables

This section presents a thorough description of the variables used for this study. An overview of the variables can be found in Appendix A: Systematic Overview of Variables – Table 1.

Dependent variables

This research measures two separate dependent variables, producing two different equations, which both appear to contribute to the innovative capacity of the firm. The first dependent variable is labeled as DV1ForwardRef, which denotes Dependent Variable 1 Number of Forward References. This first dependent variable represents the number of times the patent is referenced by other operating firms in the entire USPTO patent database, between 1990 and 2013. The second dependent variable is labeled as DV2RangeIndus, which corresponds to Dependent Variable 2 Range Industries of Patent. This second dependent variable consists of the entire range of industries in which the patent (under investigation) operates. Both dependent variables are measured on a ratio scale. The two variables tend to measure a firm’s innovative performance in different ways; while the first dependent variable tries to understand the success of the actual innovation, the second variable focuses on the range of national USPTO classes in which the patent (under investigation) actively operates. Therefore, differences across the result may exist.

Independent variables

This research will make use of three types of independent variables, labeled as IVTime, ComputeIVGeography and IVScience (previously indicated as time, geography and science). All of these independent variables make use of backward citations, and are calculated based on averages.

(25)

The first independent variable is IVTime, which represents Independent Variable Time. This first independent variable uses backward citations to determine the average number of years to which the patent under investigation refers. The time frame of the backward citations covers a total period of 24 years, ranging from 1980 until 2003. For this research, the ratio variable IVTime is collapsed into three distinct groups: IVTime Group 1 indicates all the backward citations up to three years, IVTime Group 2 indicates all the backward citations from 4 through 6 years, and IVTime Group 3 indicates all backward citations from 7 through 10 years. Following this approach, the variable explores whether the firm encodes inferences from history (old knowledge) to establish new knowledge (Levitt & March, 1988, p. 319).

The second independent variable is ComputeIVGeography and represents the Compute Independent Variable Geography. This second independent variable uses backward citations to determine the geographic area in which the patent under investigation is in effect. The computed variable knows three sub-dummy variables, namely City, State and Country. Each of these dummy variables is either assigned a value of 0 (indicating equality in City, State or Country) or a value of 1 (indicating difference in City, State or Country). Subsequently, these dummy variables are computed into the single variable ComputeIVGeography, which, again, has collapsed into three distinct groups. ComputeIVGeography Group 1 indicates all the backward citations scoring 0 or 1. ComputeIVGeography Group 2 indicates all the backward citations scoring 2 points. And finally, ComputeIVGeography Group 3 indicates all the backward citations scoring 3 points. Thus, for example, scoring 3 points ultimately means that the patent refers to a backward citation in effect in a different city, a different state and in a different country. Following this approach, the variable explores whether the firm uses distant contexts to raise the various areas of the knowledge landscape.

(26)

Finally, the third independent variable is named IVScience, which denotes the Independent Variable Science. This independent variable uses backward citations to determine the different types of industries that the patent under investigation refers to. Central to this variable is the assumption that market diversification is anticipated by technological diversification: firms that operate in different industries tend to operate different technologies (Breschi, Lissoni, & Malerba, 2003) (Pavitt, 1998). This assumption is strengthened by the dissimilar range of industries to which the patent refers, examples are D12: Transport, 5: Beds, 16: Miscellaneous hardware and 285: Pipe joints or couplings. With this assumption, this Master’s thesis chooses to follow the approach of Gambardella and Torrisi (1998). The authors measure technological diversification by the number of patents in five different sectors, such as computers and telecommunications equipment (Gambardella & Torrisi, 1998). Kodama (1986) follows a similar manner. According to him, ‘each sector’s R&D activity outside its principal product fields can be considered as technological diversification’ (Kodama, 1986, p. 291). Because of the above reasons, this Master’s thesis chooses to measure technological diversification via the different types of industries that the patent under investigation cites. The variable is measured on a ratio scale.

Control variables

This research uses five types of control variables; four of them are tested in the first equation operating on the dependent variable DV1ForwardRef, and four of them are tested in the second equation operating on the dependent variable DV2RangeIndus1.

1_{In this research, a total of six control variables were used. Yet, due to an extremely high} standard error for CV1TotPatDatab, labeled as Control Variable 1 Total Active Patents in Database per Firm, the data seemed unreliable. Hence, this variable was dropped.

(27)

CV2RangeIndus

The first control variable, CV2RangeIndus, operating under the label Control Variable 2 Range Industries of Patent, only holds as a control variable for the first equation and dependent variable DV1ForwardRef. This second control variable measures the range of industries in which the patent under investigation operates, and thus covers the same data as that of DV2RangeIndus. Note that the variable measures a different construct than the IVScience. Both are related to technological diversification, yet in a very different way. The independent variable science refers to technological diversification in that it tries to determine which types of industries the patent under investigation refers to and therefore uses for its own knowledge base. The control variable CV2RangeIndus, however, refers to the actual number of industries in which the patent under investigation operates itself (no backward referencing applied). Firms that refer to patents (by backward citations) that operate in various industries are expected to operate in a broad range of industries with their own patents, and therefore, possess a high number of forward references. Following the empirical results of Garcia- Vega (2006), it is expected that the number of patents increase with the degree of technological diversification applied by the patent of the firm. Possible explanations mainly relate to the reduction in risk from technological investments (Garcia- Vega, 2006, p. 230)

CV3ForwardRef

The third control variable is defined as CV3ForwardRef, operating under the label Control Variable 3 Number of Forward References, and only holds as a control variable for the second equation and dependent variable DV2RangeIndus. This third control variable measures the number of times the patent is referenced by other operating firms in the database, and thus covers the same data as that of DV1ForwardRef. Following Trajtenberg

(28)

measures of the social value of innovations. Patents that are often referenced, therefore, tend to be of value to the industry (Trajtenberg, 1990). The variable CV3ForwardRef measures, via the total number of references, the popularity of the patent under investigation. This Master’s thesis predicts that firms with popular patents exhibit high innovative ability, mainly due to the success of these patents.

CV4InterIndus

The fourth control variable is CV4InterIndus, labeled as Control Variable 4 Interest in Industry per Year. This fourth control variable tends to measure the amount of interest in the industry, primarily by looking at the number of filed patents each year. The variable operates as a ratio variable and holds for both the dependent variables. This Master’s thesis expects to find a positive relationship between the industry interest, as calculated by the number of filed patents each year, and the innovative ability of the firm. It is proposed that firms that have valid patents, filed in years in which the industry interest is high, tend to have strong innovative capacity. Possible explanations relate to strong market and technological knowledge, as well as strong entrepreneurial orientation (Tse, Kin, & Zheng Zhou, 2005). The timing of these firms seems crucial: the firms have managed to file patents during the most successful industry periods.

CV5SelfCitation

The fifth control variable is designated as CV5SelfCitation and labeled as Control Variable 5 Application of Self Citation. This fifth control variable shows whether the patents from the sample are actively involved in self-citation. The variable is determined as a dummy variable, with a value of 0 for firms that do apply self-citation and a value of 1 for firms that do not. The control variable holds for both the dependent variables. Following Miller et al.

(29)

(2007), this Master’s thesis expects to find a negative impact of selfcitation on the innovative performance of a firm. Possible explanations relate to the lack of knowledge expansion initiated by the firm (Miller, Fern, & Cardinal, 2007). Firms that apply selfcitation tend to be of a high age: old firms are able to use backward referencing extensively due to their long history of innovations. Also, old firms may be stuck in their old routines and maintain a high exploitative focus. This tends to influence their innovative capacity in a negative way (Sørensen & Stuart, 1999).

CV6TotFirmsIndus

Finally, the last control variable is defined as CV6TotFirmsIndus, and it carries the label of Control Variable 6 Total Number of Firms per Year. This final control variable measures the total number of firms operating in the industry each year. The variable is operated as a ratio variable and holds for both the dependent variables. This Master’s thesis predicts that firms, which publish patents in years in which industry participation is high, tend to have strong innovative capacity. The reason for this mainly relates to the concentration of the industry, for example, as Mansfield (1981, p. 612) states: a high level of industrial concentration might promote technological change in that it leads to large amounts of R&D spending. Also, high industry concentration leads to a larger share of R&D spending going for ambitious, risky projects (Mansfield, 1981, p. 612). Hall and Ham (1999) refer to these same conclusions and propose that a rise in patenting coincides with a significant entry by firms (Ham & Hall, 1999).

(30)

Validity, reliability and objectivity

To be of value, the research design must withstand tests of validity, reliability and objectivity. Validity is the ‘extent to which (a) the data collection methods accurately measure what they were intended to measure and (b) the research findings are really about what they profess to be about’ (Saunders & Lewis, 2012, p. 127). For the within-subject design it is important to focus on strong operation and measurement of the sample, mainly because the study only allows for data from a single industry. Generalization, therefore, must be handled with care. Reliability is defined as ‘the extent to which a measurement is consistent and free from error, when used by the same rater (intra-rater reliability), or when used by different raters (inter-rater reliability)’ (Barret, McCreesh, & Lewis, 2014, p. 11). This study aims to provide strong consistency across results, both valid and reliable, by studying a sample of raw data. Finally, through objectivity, the researcher must ensure a value-free, unbiased research study. Accepting competing theories and providing criticism will increase the objectivity of this study.

(31)

Results

This section presents the results of the research conducted. First, descriptive statistics introduce the independent, dependent and control variables. Subsequently, analysis with respect to normal distribution will be executed for the same kinds of variables, after which amendments will maximize the value of the data (outlier analysis). Next, bivariate correlation analysis indicates the most important, significant correlations. Finally, hierarchical multiple regressions are performed in order to test the proposed hypotheses of non-local search and firm innovation. It is important to mention the two-folded operation of some of these steps; some are operated on the granted patents and some on the filed patents.

Descriptive statistics

The current research tends to make use of eleven variables, either labeled as independent, dependent or control (see variable overview in section Appendix A: Systematic Overview of Variables – Table 1). The descriptive statistics of the continuous variables are outlined in Table 2: Descriptive Statistics for Continuous Variables – Granted and Table 4: Descriptive Statistics for Continuous Variables – Filed, while the frequencies of the dummy variables are outlined in Table 3: Frequencies for Dummy Variables – Granted and Table 5: Frequencies for Dummy Variables – Filed. These tables can be found in Appendix B: Descriptives.

With respect to the database of granted patents, a total number of N=103 patents seem to be valid for operation. Overall, it can be seen that the industry peaked between the years 1990 and 1993, with 56 active patents. Leading firms were Everest & Jennings, Inc., Invacare Corporation and Kurt Manufacturing Company, all operating in the United States. The

(32)

variables that might have similar issues are the identical variables CV3ForwardRef and DV1ForwardRef, and they may therefore need adjustment. Some of the choices of the firms seem remarkable; firms seem to actively use backward citations of patents dating back four to six years (IVTime Category 2: 56.3%), operating in approximately two different industries (IVScience =1.6424) and in two different spaces (ComputeIVGeography Category 2: 49.5%). Second, it seems that the related innovative performance of firms explained by the amount of reference is approximately eight (DV1ForwardRef =7.91045), and the innovative performance explained by operating industries is approximately two (DV2RangeIndus =2.1262).

With respect to the database of filed patents, a total number of N=81 patents seem to be valid for operation. Again, the years 1990 through 1993 seem to be very important for the industry; 43 patents were active. Leading firms were Invacare Corporation, Kurt Manufacturing Company and Medical Composite Technology, all operating in the United States. Table 4 shows similarities regarding the above findings: a high standard deviation for the variable CV1TotPatDatab and the particular use of patents dating back four to six years (IVTime Category 2: 56.8%), operating in approximately two different industries (IVScience =1.4447) and in two different spaces (ComputeIVGeography Category 2: 48.1%). The related innovative performance of firms explained by the amount of reference is approximately eight (DV1ForwardRef y =8.2099), and the innovative performance explained by operating industries is approximately two (DV2RangeIndus = 2.0617).

Outlier Analysis

Before deciding whether the variables seem normally distributed, a thorough assessment regarding outliers is needed. Outliers are data points or scores that are different from the remainder of the scores, and which tend to incorrectly influence conclusions

(33)

regarding the data (Pallant, 2005, p. 250). They tend to extend more than 3 box-lengths from the edge of the box of normal data, and therefore often require transformation or deletion. The transformation or deletion of data is only executed when necessary, mainly due to the belief that more data improves the actual analysis and findings. The associated tables and graphs can be found in Appendix C: Outlier Analysis.

Looking at the Mean and 5% Trimmed Mean values of the individual boxplot in Appendix C: Granted Outlier Analysis – Table 6 (and accompanied boxplots Graphs 1 to 3), it seems that the variables IVTime, ComputeIVGeography, IVScience, CV2RangeIndus, CV5SelfCitation and CV6TotFirmsIndus do not indicate extreme negative or positive scores which influence the mean value of the variables2. As seen in the Descriptives section, the control variable CV1TotPatDatab has an extremely high standard deviation. Subsequently, the extreme values of the variable are very high or low ( =1201.9223, 5% Trimmed =295.6036). The variable will therefore not be used. While analyzing the findings, four other variables seem ready for transformation: the identical variables CV3ForwardRef and DV1ForwardRef ( =7.6109, 5% Trimmed =6.7497) and the identical variables CV2RangeIndus and DV2RangeIndus ( =2.1262, 5% Trimmed =1.9984). The first two variables are in need of adjustment, mainly due to a high difference in means; the select-case-if option withholds all findings of CV3ForwardRef and DV1ForwardRef > 24. Additionally, though less necessary, the second two variables show one extra ordinary variable of value 8.

The select-case-if option therefore withholds all findings of CV2RangeIndus and

DV2RangeIndus > 5. It is important to mention that the variable TotalCit must always be > 0,

(34)

a patent must refer to other citations in order to deliver data valuable for the analysis (this holds for both databases).

Following the exact same reasoning, Appendix C also highlights figures regarding the Filed Outlier Analysis – Table 7 (and accompanied boxplots Graphs 4 to 6). Again, the control variable CV1TotPatDatab ( =861.1852, 5% Trimmed =134.4438) is not operated in this research due to an extreme standard error, and four variables are eligible for transformation. These seem to be equal to the variables stated above, namely the identical variables CV3ForwardRef and DV1ForwardRef ( =8.2099, 5% Trimmed =7.3086) and the identical variables CV2RangeIndus and DV2RangeIndus ( =2.0617, 5% Trimmed =1.9170). The first two variables are again in need of adjustment, mainly because of a high difference in means; the select-case-if option this time withholds all findings of CV3ForwardRef and DV1ForwardRef > 29. Additionally, the second two variables show only one extraordinary variable of value 8, and transformation of the two variables withholds all findings of

CV2RangeIndus and DV2RangeIndus > 5.

After these amendments, the granted database covers a total number of N=95 patents, while the filed database covers a total number of N=74 patents. Again, it is important to mention that transformation or deletion of the data is only executed when the outlier influences the mean value significantly. This is mainly due to the belief that more data improves the validity of the research.

Normal Distribution Analysis

Subsequently, normal distribution of the variables is tested in three distinct ways; via the skewness and kurtosis, via the Shapiro-Wilk Test, and via the graphical interpretation of the histogram and normality plot. The associated tables and graphs can be found in Appendix D: Normality Analysis.

(35)

First, in order to test for normality, analysis upon the skewness and kurtosis error is performed. Whereas skewness provides information regarding the symmetry of the distribution, kurtosis tends to focus more on the ‘peakedness’ of the distribution (Pallant, 2005, pp. 51- 52). A perfectly normal distribution would result in a value of 0 for both skewness and kurtosis, indicating a symmetric, bell-curved distribution. The observations in Table 8: Skewness and Kurtosis – Granted indicate that all variables tend to be not normal, given their values for skewness and kurtosis. The table shows how the variables IVTime (Skew=-.097) and CV4InterIndus (Skew=-.064) contribute to an almost normal skew. Additionally, the variable IVScience exhibits an almost normal kurtosis (Kur=0.05). Table 9: Skewness and Kurtosis – Filed shows, again, how the variable IVTime exhibits an almost normal skew (Skew=-.067). Overall, both databases show that the variables tend to be rather flat and clustered negatively or positively.

Second, the Kolmogorov-Smirnov statistic tends to assess the normality of the distribution of scores. More specifically, a non-significant result (note Sig<0.05) indicates normality of the operated variable (Pallant, 2005). For this research, the Shapiro-Wilk Test (W) is used. This test is comparable to the Kolmogorov-Smirnov test, yet more appropriate for small sample sizes (N<2000) (IBM, 2011). While observing both the granted and filed database variables in Appendix D: Tables 10 and 11, none of them seems to be significant (granted database: Sig<0.02; filed database: Sig<0.00). Again, this finding concludes that all variables operate in a rather unique way.

Third, graphical interpretation by means of histograms and normality plots is used to assess the normal distribution of the variables in a more subjective way (Appendix D: Normality Analysis Graph 7-14). The histograms of first dependent variable DV1ForwardRef and identical control variable CV3ForwardRef, as well as independent variable IVScience, are

(36)

from the normal distribution in the histogram or from the straight fit line of the normality plots, does not occur at any of the variables.

Bivariate Correlation Analysis

Appendix E: Bivariate Correlation Analysis outlines all of the results from the bivariate correlation analysis. Both the Spearman Rank correlations (rho), as well as the Pearson product-moment correlations (r) are presented, mostly because of the unique distribution of the variables (as indicated in the section Normality Analysis). The use of Pearson product-moment correlations tends to be preferred over the use of Spearman Rank correlations, since these correlations have the highest ability to detect differences and relationships (Pallant, 2005, p. 82). The most essential correlations, which tend to acknowledge a significant effect (Sig<0.05, and marked with an asterix*), will be discussed. Especially, the effect between the independent and dependent variables, moderated with the control variables, is of most importance. Each dependent variable knows a total number of four control variables: a match from CV2RangeIndus, CV3ForwardRef, CV4InterIndus, CV5SelfCitation and CV6TotFirmsIndus, respectively.

DV1ForwardRef

First, the correlations regarding the dependent variable DV1ForwardRef are analyzed for both the granted and filed databases. Correlations can be found in the Appendix E: Bivariate Correlation Analysis, specifically Tables 13 and 14. It seems that for the granted database, either with or without operation of the control variables, no significant effects regarding the independent variables are found. This changes when attention is aimed at the filed database (Tables 16 and 17); the independent variables ComputeIVGeography (r=-.258; Sig=.026) and IVScience (r=.254; Sig=.029) seem to have a significant negative and a

(37)

significant positive effect on the dependent variable DV1ForwardRef. Though, with the introduction of the control variables, these significant effects seem to evaporate (ComputeIVGeography: r=-.205; Sig=.088) (IVScience: r=.086; Sig=.478).

DV2RangeIndus

The second dependent variable shows more interesting correlations (Appendix E: Bivariate Correlation Analysis, specifically Tables 17 and 19). The granted database indicates a significant correlation between the IVScience and DV2RangeIndus (r=.482; Sig=.000). Even after introduction of the control variables, the effect between the variables seems to retain significance (r=.381; Sig=.000). The relationship between the IVTime and dependent variable seems to be interesting; while introducing the control variables, the relationship between the IVTime and DV2RangeIndus becomes significant (r=.281; Sig=.007). As for the filed database, the independent variables ComputeIVGeography (r=-.258, Sig=.027) and IVScience (r=.572; Sig=.000) indicate a significant negative and a significant positive effect upon the dependent variable DV1ForwardRef. Again, after the introduction of the control variables, the effect between the IVScience and dependent variable still holds (r=.366; Sig=.002).

Other

Other interesting correlations stem from the relationship between the two dependent variables, DV1ForwardRef and DV2RangeIndus. The relation is significant, possessing a rather high positive coefficient (r=.399; Sig=.000). This indicates how both variables tend to measure the same type of underlying construct of innovation. Subsequently, CV6TotFirmsIndus exhibits a positive, significant relation with DV2RangeIndus (r=.286;

(38)

participation is high, tend to denote a high level of firm innovation. Finally, the relation between CV4InterIndus and CV6TotFirmsIndus is positively significant (r=.671; Sig=.000). Firms that perform much patent activity tend to operate in the years in which industry participation is high.

Hierarchical Multiple Regression Analysis

This section provides evidence regarding the four hypotheses established at the start of this Master’s thesis (details are shown in Appendix F: Hierarchical Multiple Regression Analysis). Providing evidence requires the introduction of hierarchical multiple regression analysis, in other words, regression analysis that enters the variables in a specific step-wise order, mainly to control for some variables. Due to the more interesting findings for the second variable DV2RangeIndus operating in the granted database, it is decided to pay attention to these related results. The results from regression analysis of DV1ForwardRef for both the granted and filed database and of DV2RangeIndus for the filed database can be found in Appendix G: Results Regression Analysis Other Equations. In addition to the results, this section will give attention to related assumptions, such as multicollineairity, homoscedasticity and linearity. Finally, an additional time frame analysis is presented to analyze whether the effects hold in time.

The database contains a total of 103 patents. After previous transformation, the final sample is set at 95 patents, mostly due to the outlier analysis. The hierarchical multiple regression analysis is carried out in terms of five scenarios.

First, the control variables are introduced, corresponding to Model 1 in the regression output of Appendix F. Following the outcomes stated in Table 20 and 21, Model 1 presents the effect of control variables CV3ForwardRef, CV4InterIndus, CV5SelfCitation and

(39)

CV6TotFirmsIndus upon the dependent variable DV2RangeIndus. The results show that more than 30% of all variance was explained by the control variables (R Square=.306; Sig=.000). The model was significant on a 5% level (F=9.916; Sig=.000). The contribution of the individual control variables appeared to be significant for both CV3ForwardRef (B=0.414; Sig=.000) and CV4InterIndus (B=.365; Sig=.003).

Model 2 introduces the independent variable IVTime to the first model. The first hypothesis, the greater the degree of search in time, the greater the related innovative performance, is tested by means of this second model. Again, following the results of Tables 20 and 21, the model seems to explain 36.1% of all variance, an increase of 5.5% (R Square=.361, R Square Change=.055). The model was held significant on a 5% level (F=10.036; Sig=.000). This also holds true for the contribution of the independent variable IVTime (F Change=7.606; Sig=.007). The regression coefficient for IVTime was, again, significant and provides support for Hypothesis 1 (B=.240; Sig=.007).

Model 3 introduces the independent variable ComputeIVGeography to the first model of control variables. The second hypothesis, the greater the degree of geographic search, the greater the related innovative performance, is tested by means of this third model. The model seems to explain 31.7% of all variance, which reflects an increase of 1.1% (R Square=.317, R Square Change=.011). The model was held significant on a 5% level (F=8.270; Sig=.000). However, the contribution of the variable IVTime was not significant (F Change=1.477; Sig=.227). This also holds for the regression coefficient of the variable (B=-.114; Sig=.227). Hypothesis 2 is therefore not supported.

Model 4 introduces the independent variable IVScience to the first model of control variables. The third hypothesis, the greater the degree of search in science, the greater the related innovative performance, is tested by means of this model. The model seems to explain