Cooperation in non-R&D activities: the influence on the
product innovation performance of Dutch manufacturing
firms
A mixed-methods study about the influence of non-R&D-cooperation on the product innovation performance of Dutch manufacturing firms
Master thesis Business Administration – Strategic Management
Nijmegen School of Management
Jean-Paul Latour
S1025147
Supervisor: Dr. P.M.M. Vaessen
Second examiner: Dr. K.F. van den Oever
Date: 15 June 2020
Abstract
It has been found that many firms are innovative without having an own R&D-department. Based on the endogenous growth theory, a possible explanation for this is that firms without an R&D-department become innovative through cooperation with other firms because that might result in the spillover of knowledge. Therefore, this thesis seeks to explain to what extent non-R&D-cooperation between manufacturing firms results in knowledge spillovers contributing to their product innovation performance. Additionally, the influence of spatial proximity between cooperation partners and the influence of a firm’s absorptive capacity on this relationship have been investigated. In order to answer the research questions, a mixed-methods study has been conducted. For the quantitative analysis, data from the 2015 European Manufacturing Survey has been used. For the qualitative analysis, five interviews have been held with employees of five different firms. The results have made clear that non-R&D-cooperation has no significant effect on a firm’s product innovation performance. Moreover, it has been found that spatial proximity between cooperation partners has no significant influence on this relationship. Additionally, the qualitative analysis has revealed that the absorptive capacity of a firm has an influence on the amount of knowledge that spills over due to non-R&D-cooperation. Finally, the qualitative analysis has made clear that there are more possible explanations for the innovativeness of firms without an own R&D-department. However, future research is necessary to generalize these findings.
Key words: non-R&D-cooperation, spatial proximity, absorptive capacity, product innovation, manufacturing industry
Preface
With the completion of this thesis I am finishing the master program Business Administration, specialisation Strategic Management. The title of this thesis is: “Cooperation in non-R&D activities: the influence on the product innovation performance of Dutch manufacturing firms.” To finish this thesis, I have made many efforts and sometimes stressful times have appeared. Moreover, it was a unique period due to the corona virus. It was very weird to me that the University was closed and that all conversations took place via Skype. However, I am proud that I have been able to finish my master thesis and I am very thankful to the people who helped me to achieve this.
First, I would like to thank my supervisor Peter Vaessen for his support during the whole process. He gave me many tips to improve my thesis. In addition, I also appreciate the time that my second examiner, Koen van den Oever, made available to provide me with valuable feedback on my research proposal.
Moreover, I would like to thank the five respondents as well as their companies for their willingness to participate in an interview. It gave me more in-depth insights into the quantitative findings. Unfortunately, due to the corona virus I was only able to visit one of the five companies, but I am very grateful that the other respondents were able to have a phone conversation with me.
Finally, I would like to thank my family and friends for their moral support during the whole process.
I hope you enjoy reading my master thesis.
Content
Chapter 1 – Introduction ... 1
Chapter 2 – Theoretical framework ... 4
2.1 Introduction ... 4
2.2 The dependent variable innovation defined ... 4
2.3 The independent and moderating variables defined ... 6
2.3.1 Knowledge spillovers generated from non-R&D-cooperation ... 6
2.3.2 Tacit and explicit knowledge ... 7
2.3.3 Spatial proximity ... 7
2.3.4 Absorptive capacity ... 8
2.4 Linking the concepts ... 9
2.4.1 Non-R&D cooperation and product innovation ... 9
2.4.2 The moderating effect of spatial proximity ... 10
2.4.3 The moderating effect of absorptive capacity ... 12
2.5 Conceptual model ... 13
Chapter 3 – Methodology ... 15
3.1 Introduction ... 15
3.2 Research method ... 15
3.3 Quantitative analysis ... 15
3.3.1 Operationalisation of the quantitative analysis ... 16
3.3.2 Method for analysing the quantitative data ... 18
3.3.3 Validity and reliability of the quantitative analysis ... 19
3.4 Qualitative analysis ... 19
3.4.1 Operationalisation of the qualitative analysis ... 20
3.4.2 Methods for analysing the qualitative data ... 20
3.4.3 Validity and reliability of the qualitative analysis ... 20
3.5 Research ethics ... 20
Chapter 4 – Data analysis ... 22
4.1 Introduction ... 22
4.2 Sample characteristics ... 22
4.3 Variable construction ... 24
4.3.1 Construction of the dependent variable ... 24
4.3.2 Construction of the independent variable ... 24
4.3.3 Construction of the moderating variables ... 25
4.3.4 Construction of the control variables ... 25
4.4 Univariate analysis ... 26
4.5 Bivariate analysis ... 28
4.6 Multivariate analysis ... 31
4.7 Qualitative analysis ... 37
Chapter 5 – Conclusion and discussion ... 45
5.1 Introduction ... 45
5.2 Conclusion ... 45
5.3 Discussion ... 47
References ... 51
Appendices ... 56
Appendix I Interview questionnaire ... 57
Appendix II Tree structures interviews ... 60
Appendix III SPSS Output skewness and kurtosis transformations % highly educated employees ... 62
Appendix IV Assumption check binary logistic regression analysis ... 63
Appendix V Relevant output binary logistic regression analysis ... 66
Appendix VI Output relationship R&D-cooperation on product innovation with non-R&D-cooperation as moderator ... 74
Appendix VII General information interviewed firms ... 76
Appendix VIII Interview transcripts ... 77
Appendix IX Coded interviews ... 105
Appendix X Signed Research Integrity Form ... 129
List of Tables and Figures Tables
Table 1 Operationalisation table 16
Table 2 Firm size of companies participating in the EMS 22
Table 3 Overview type of industry 23
Table 4 Cronbach’s Alpha if item deleted non-R&D-cooperation 25 Table 5 Frequencies of product innovation (dependent variable) 26
Table 6 Descriptive statistics of metric variables 27
Table 7 Old and new skewness and kurtosis variable highly educated employees 27 Table 8 Descriptive statistics of metric control variables 28
Table 9 Correlations of the bivariate analysis 29
Table 10 Results binary logistic regression analysis 34
Table 11 Overview outcomes of the hypotheses 37
Figures
Chapter 1 – Introduction
Chapter 1 consists of the introduction to the research topic. Also, the research objective, research question and relevancy are described. The chapter ends with an outline of the remainder of this thesis.
Firms need to be innovative to deal with the rapidly changing environment (Aldieri, Sena, & Vinci, 2018; Luiz Fernando de Paris Caldas, 2019). In most economic literature knowledge generated by Research and Development (R&D) is seen as the key driver to become innovative (Barge-Gil, Jesús Nieto, & Santamaría, 2011; Brouwer & Kleinknecht, 1997; Lopez-Rodriguez & Martinez-Lopez, 2017). However, the European Community Innovation Survey (CIS-3) has found for 15 countries in Europe that almost half of the European firms considered to be innovative did not have an own R&D department (Lopez-Rodriguez & Martinez-Lopez, 2017). A possible explanation for this finding is that knowledge generated by a firm’s R&D-department also generates opportunities to become innovative for firms without an own R&D department (Acs, Braunerhjelm, Audretsch, & Carlsson, 2009). According to the endogenous growth theory, these opportunities may arise due to the spillover of knowledge between firms through cooperation in non-R&D activities (Acs et al., 2009; Grossman & Helpman, 1991). However, there is actually less evidence for this suggestion. Therefore, this research will extent the current literature by finding out if non-R&D-cooperation between manufacturing firms actually results in knowledge spillovers contributing to their innovation performance.
Knowledge spillovers1 are defined as unintended flows of costless knowledge generated by
informal means and used by a recipient firm for its own innovation purposes (Alcácer & Chung, 2007; Arrow, 1962; Cohen & Levinthal, 1990). Regarding innovation, there are different types. In the literature, it is argued that an orientation with a focus on new knowledge, skills and processes contributes to product innovation (Lee, Lee, & Garrett, 2019). Due to the focus of this research on knowledge spillovers, product innovation is the type of innovation that will be investigated. A more detailed explanation for this choice will be given in chapter 2. Moreover, it seems that the extent to which non-R&D-cooperation between firms results in knowledge spillovers contributing to a firm’s product innovation performance is dependent on two factors: the spatial proximity between cooperation partners and the absorptive capacity of the focal firm.
There is empirical evidence that knowledge spillovers are geographically bounded within the region where the new knowledge is created (Audretsch & Feldman, 2004; M. Porter, 1990). This is for example shown by Jaffe, Trajtenberg, and Henderson (1993), they found that inventors are more likely to cite patents of other inventors who are in their geographical proximity. However, the current literature pays attention to virtual teams. Virtual teams are defined as groups of interdependent workers who are geographically dispersed and electronically dependent on each other, through the use of electronic tools like Skype or knowledge repositories (Gibson & Gibbs, 2006). Exciting about virtual teams is that they allow people worldwide to use the most relevant knowledge (Kozlowski, Kirkman, Gibson, & Kim, 2012). The emergence of virtual teams raises the question if spatial proximity between cooperation partners is still important. Moreover, also Boschma (2005) and Beugelsdijk and Cornet (2001) found no evidence for the importance of spatial proximity between cooperation partners. This study will narrow these contradictory thoughts in the literature by investigating the influence of spatial proximity between cooperation partners on the relationship between non-R&D-cooperation and product innovation.
The absorptive capacity of a firm is the second aspect that influences the extent to which knowledge spills over through non-R&D-cooperation. Absorptive capacity is one of the most influential concepts in the management and innovation literature (Aldieri et al., 2018). It is introduced by Cohen and Levinthal (1990) and refers to the ability of a firm to identify, record and exploit knowledge from the environment. It is suggested that a firm’s absorptive capacity is largely a function of prior knowledge because prior knowledge determines the ability to recognize the value of new information (Cohen & Levinthal, 1990). To strengthen the literature about the concept absorptive capacity, this study will take into account the influence of a firm’s absorptive capacity on the relationship between non-R&D-cooperation and product innovation.
Hence, the purpose of this research is not only to give insight in the extent to which non-R&D-cooperation results in the spillover of knowledge that contributes to the product innovation performance of manufacturing firms. The purpose is also to investigate the influence of spatial proximity between cooperation partners and the influence of the absorptive capacity of the focal firm on this relationship. To achieve this purpose, the following research question is set up:
“To what extent does non-R&D-cooperation between manufacturing firms result in the spillover of knowledge that contributes to their product innovation performance?”
In addition, the following sub research questions are set up:
a) To what extent does spatial proximity between non-R&D-cooperation partners influence the relationship between non-R&D-cooperation and product innovation? b) To what extent does the absorptive capacity of the focal firm influence the relationship
between non-R&D-cooperation and product innovation?
This study has several contributions, both for practice and science. The practical relevance of this research is that it gives insight to manufacturing firms in the extent to which non-R&D-cooperation contributes to their product innovation performance. Furthermore, it will find out to what extent spatial proximity and absorptive capacity are important aspects to take into account to benefit from non-R&D-cooperation. Regarding the scientific relevance, there is currently less evidence about the influence of non-R&D-cooperation on product innovation. This research will strengthen the literature about this topic. Moreover, in most of the literature it is assumed that spatial proximity between cooperation partners is important, this research will find out if this is still the case. Also, this study could give new scientific insights in the influence of a firm’s absorptive capacity on the relationship between non-R&D-cooperation and product innovation. Furthermore, chapter 2 starts with an extensive review of the existing literature regarding the core concepts of this research. The last reason for the relevance of this study is that a combination of quantitative and qualitative research has seldom been used in current studies about the concept of knowledge spillovers.
The remainder of this thesis is structured as follows: chapter 2 discusses the theoretical background; it provides an outline of relevant definitions and the conceptual model is demonstrated. Chapter 3 provides the methods that are applied during this research. After that, in chapter 4 the data analysis phase will be discussed. This chapter has two separate parts: the quantitative analysis and the qualitative analysis. Finally, in chapter 5 the research questions are answered. Also, the discussion of this research and the research limitations and directions for future research are shown in chapter 5.
Chapter 2 – Theoretical framework 2.1 Introduction
In this chapter, the theoretical framework will be described. First, the core concepts of this research are defined whereby in section 2.2 the dependent variable is defined and demarcated and in section 2.3 the independent and moderating variables are defined and demarcated. After that, in section 2.4 the core concepts are linked to each other resulting in hypotheses that are tested during the empirical phase in chapter 4. Finally, the chapter ends in section 2.5 with a representation of the conceptual model in which the hypotheses are merged.
2.2 The dependent variable innovation defined
Innovation is seen as one of the most important sources of sustainable competitive advantage in today’s fast changing environment because it leads to product and process improvements and therefore helps firms to grow and survive (Tojeiro-Rivero & Moreno, 2019). Furthermore, innovating firms are generally more profitable than non-innovating firms (Atalay, Anafarta, & Sarvan, 2013). The term innovation stems from the Latin expression ‘innovation’, which means “to create something new” (Lorenz, 2010).
In the literature, there is a wide range of definitions findable attempting to explain what innovation actually means (Lorenz, 2010). A definition that is extensively used is described in the OECD and Communities (2005, p. 46): “An innovation is the implementation of a new or significantly improved product (good or service), or process, a new marketing method, or a new organisational method in business practices, workplace organisation or external relations.” A common feature mentioned by OECD and Communities (2005) is that an innovation must have been implemented. When an innovation is not implemented or commercialized, than it is called an invention but not an innovation (Lorenz, 2010).
The definition discussed in the previous paragraph distinguishes four main types of innovations: product innovations, process innovations, marketing innovations and organizational innovations. These types are widely presented in the literature (Kahn, 2018; Lee et al., 2019; Lorenz, 2010; OECD & Communities, 2005). In the Oslo Manual, separate definitions for each type of innovation are discussed.
These separate definitions are:
• Product innovation: the introduction of a good or service that is new or significantly improved with respect to its characteristics or intended uses (OECD & Communities, 2005, p. 48)
• Process innovation: the implementation of a significantly improved production or delivery method (OECD & Communities, 2005, p. 49)
• Marketing innovation: the implementation of a new marketing method (OECD & Communities, 2005, p. 49)
• Organizational innovation: the implementation of a new organisational method (OECD & Communities, 2005, p. 51)
Although innovation is a broad term and consists basically of four types, these four types can be further distinguished between technological and non-technological innovations. Product and process innovation are seen as technological innovation and marketing and organizational innovation are seen as non-technological innovation (Phillips & Phillips, 1997). Technological innovation is about the development of new products and new production techniques (OECD & Communities, 2005) whereas non-technological innovations are more specific to a firm, for example through new organizational concepts (Schmidt & Rammer, 2007). Technological innovation has a crucial impact on a firm’s sustainable competitive advantage (Weihong, Caitao, & Dan, 2008). In contrast, non-technological innovation has only small effects on a firm’s profit margin (Schmidt & Rammer, 2007). Due to the focus of this research on innovation that results in sustainable competitive advantage, only technological innovation will be taken into account.
Zooming in on technological innovation, research has shown that an orientation which focuses on new knowledge, skills and processes helps firms to introduce product innovations (Lee et al., 2019). An orientation focusing on the utilization of existing resources and increasing efficiency augments the likelihood of process innovation (Lee et al., 2019). The focus of this study is on the gaining of new knowledge generated from knowledge spillovers which results in the conclusion that product innovation is the most appropriate type of innovation to measure. Therefore, this research is demarcated to the effect of knowledge spillovers generated from
improvements in technical specifications, components and materials, incorporated software, user friendliness or other functional characteristics (OECD & Communities, 2005, p. 48).”
2.3 The independent and moderating variables defined
In the previous section, the dependent variable is described and defined. This section will demarcate the interpretations of the independent and moderating variables.
2.3.1 Knowledge spillovers generated from non-R&D-cooperation
It may be argued that knowledge is a firm’s most important asset for achieving innovations (Grant, 1996). Recent literature has shown that external sources of knowledge are just as important as internal knowledge sources (Cassiman & Veugelers, 2006; Zook & Rigby, 2002). Gaining external knowledge is seen as less costly and it gives more certainty because it does not demand large and specific investments (Audretsch & Feldman, 2004; Barge-Gil et al., 2011). A way to gain external knowledge is through non-R&D-cooperation because that might result in the spillover of knowledge. The concept of knowledge spillovers is seen as a central element of innovation theories (Beugelsdijk & Cornet, 2001). Some people view knowledge spillovers as leaks, but in reality they are a crucial ingredient for economic growth (Romer, 1990) because they can result in innovation (Phene & Tallman, 2014).
Knowledge has some of the characteristics of public goods. It is widely considered to be a partially excludable and non-rivalrous good (Romer, 1990). Non-rivalry suggest that a novel piece of knowledge can be used many times and in different circumstances without reducing their value (Fischer & Varga, 2003). Therefore, knowledge can spillover from one firm to another. This can be realised through non-R&D-cooperation because when companies have conversations together, these conversations might result in the spillover of knowledge. With non-R&D-cooperation is meant all informal collaborations between firms, for example cooperation on purchasing, sales/distribution, service and/or production. It is stated that innovation often occurs through combining knowledge generated from different knowledge fields (Pyka, 2000).
The literature highlights many different definitions for the concept knowledge spillovers (Acs et al., 2009; Jaffe et al., 1993; Phene & Tallman, 2014). Cohen and Levinthal (1989, p. 571) define knowledge spillovers as “any original, valuable knowledge generated in the research process which becomes publicly accessible, whether it be knowledge fully characterising an
innovation, or knowledge of a more intermediate nature.” That knowledge becomes publicly accessible means that it is generated costless. Additive to this definition, a lot of literature states that knowledge spillovers are unintended and generated by informal means (Alcácer & Chung, 2007; Arrow, 1962; Cohen & Levinthal, 1990). Furthermore, R&D cooperation only plays a minor role as a medium for knowledge spillovers (Fritsch & Franke, 2004).
In conclusion, in this research knowledge spillovers are defined as any original, costless generated knowledge that is taken from another firm in an unintended way and by informal means (through non-R&D-cooperation) and contributes to the innovation performance of the recipient firm.
2.3.2 Tacit and explicit knowledge
Knowledge can be divided into tacit and explicit knowledge (Döring & Schnellenbach, 2006; Fischer & Varga, 2003). Tacit (implicit) knowledge is defined as “knowledge that is difficult to articulate” (Baumard, 1999). Accordingly, tacit knowledge refers often to practice-based knowledge like expertise, know-how and professional intuition linked to individual capabilities (Virtanen, 2010). Explicit knowledge is the opposite of tacit knowledge. Explicit knowledge is codified, objective knowledge that is obvious to share in numbers and words (Virtanen, 2010). Moreover, explicit knowledge can be formalized and written down. Tacit knowledge is by definition not codifiable and impossible to formalize and written-down (Audretsch & Feldman, 2004). It is stated that knowledge is of little advantage when it is not shared with other organisational members (Kikoski & Kikoski, 2004).
2.3.3 Spatial proximity
This research pays attention to the influence of spatial proximity between a focal firm and their cooperation partner. It will be tested if there exists a difference in the amount of knowledge that spills over between partners who are located near to the focal firm and partners who are located further away. In the next section, a hypothesis about this subject is given based on an extensive literature review.
difference between partners located less than 50 kilometres from the focal firm, partners located more than 50 kilometres from the focal firm and partners who are located abroad. Therefore, in this research spatial proximity will be measured in kilometric distance between a focal firm and their cooperation partners, which is a reliable measure for the construct spatial proximity (De Jong & Freel, 2010).
2.3.4 Absorptive capacity
Another factor often associated with knowledge spillovers is the concept of absorptive capacity. Cohen and Levinthal (1990) stated that the ability to recognize the value of new information, assimilate it and apply it to commercial ends constitute a firm’s absorptive capacity. Given the increasingly important role of external knowledge flows, absorptive capacity has become a key driver for a firm’s competitive advantage (Cockburn & Henderson, 1998). A firm’s absorptive capacity depends on their existing stock of knowledge, which is embedded in its products, processes and people (Cohen & Levinthal, 1990; Escribano, Fosfuri, & Tribó, 2009).
The absorptive capacity of a firm is often measured by the firm-level R&D investments (Cohen & Levinthal, 1989). However, as discussed in chapter 1, past research has discovered that many firms are innovative without having an own R&D department. Therefore, this indicator for absorptive capacity seems not appropriate during this research. Moreover, more recent studies have argued that an organisation’s prior knowledge base depends on the individual units of knowledge available within the organisation (L. Kim, 2001). Accordingly, they have focused on human capital involved in the organisation (Murovec & Prodan, 2009). More specifically, these studies have used the educational level of employees as indicator for the absorptive capacity of organizations (Clausen, 2013; Giuri & Mariani, 2013; Minbaeva, Pedersen, Björkman, Fey, & Park, 2014) Therefore, in this study the absorptive capacity of a firm will be measured with the variable percentage of highly educated employees.
Last, it is important to remark that the literature mentioned two detached roles of absorptive capacity taken into account external knowledge (Cohen & Levinthal, 1990; Zahra & George, 2002). The first role is potential absorptive capacity, this includes knowledge acquisition and assimilation capabilities (Zahra & George, 2002). The potential absorptive capacity supports the firm by identifying more available knowledge flows (Escribano et al., 2009). The second role is realized absorptive capacity. This refers to knowledge transformation and exploitation (Zahra & George, 2002) and is the degree to which a firm derives benefits from external
knowledge flows (Escribano et al., 2009). In conclusion, absorptive capacity contributes in two ways to a firm’s ability to benefit from external knowledge. The first way is that firms can identify more external knowledge and the second way is that they can exploit the generated external knowledge more efficiently (Escribano et al., 2009).
2.4 Linking the concepts
This section is a continuation on the descriptions of the core concepts because it will link them. Each subsection contains a link between two core concepts. Each linkage will result in a hypothesis which is shown at the end of the subsection.
2.4.1 Non-R&D cooperation and product innovation
The first linkage is between non-R&D-cooperation and product innovation. Many authors have argued that a significant group of firms develop innovations without performing R&D activities by themselves (Aldieri et al., 2018; Audretsch & Feldman, 2004; Barge-Gil et al., 2011; Brouwer & Kleinknecht, 1997; Lopez-Rodriguez & Martinez-Lopez, 2017). It might be that non-R&D-cooperation results in knowledge spillovers that contributes to product innovation. The little evidence that has currently been found for this suggestion is discussed in this section.
The first evidence is found by a research of Fernández Sastre and Vaca Vera (2017). They used data containing information of 2815 firms in the period 2009-2011 and found that non-R&D-cooperation positively affects the introduction of new products within a company. Another example is provided by Belderbos, Carree, and Lokshin (2004) using data on a sample of Dutch innovating firms from the Community Innovation Survey (CIS) in 1996 and 1998. They found that cooperating firms are more engaged in innovation activities then non-cooperating firms. Their argument for this finding was that knowledge spills over between firms when they cooperate with each other. More evidence is found by Ahuja (2000) by conducting a longitudinal study about firms in the international chemical industry. Support was found for the hypothesis that direct ties among firms have a positive influence on a firm’s innovation output. In addition, Robinson and Stubberud (2011) used data from the Eurostat Community Innovation Survey to investigate what the most important innovation sources are for businesses in Norway. They discovered that most firms claim that external partners are important innovation sources.
Non-R&D-cooperation can occur in different fields, this research takes into account the following fields: purchasing, production, sales/distribution and service. The founder of the innovation-economy, Joseph Schumpeter, described innovation as ‘Neue Kombinationen,’ what means that successful innovations are usually the result of a combination of knowledge from different fields (Jacobs, 2006). This is in line with the findings of Zheng, Zhang, and Du (2011), they found that the use of different cooperation fields will facilitate a combination of knowledge what increases the chance that it results in innovations. Therefore, by analysing non-R&D-cooperation, the number of non-R&D-cooperation fields a firm is cooperating in will be used as independent variable.
To conclude, different studies have found empirical evidence for the positive influence of non-R&D-cooperation between firms on their product innovation performance. These findings are in line with the assumption of the endogenous growth theory that knowledge spills over between firms through cooperation (Acs et al., 2009). Furthermore, it is expected that cooperating in different cooperation fields results in a combination of knowledge that enables a firm to complete an innovation more successfully. This together results in the following hypothesis (H1):
Hypothesis 1: the number of non-R&D-cooperation fields a firm is cooperating in has a positive significant effect on product innovation.
2.4.2 The moderating effect of spatial proximity
An ongoing debate in the literature is whether spatial proximity between cooperation partners is important for the spillover of knowledge. This section will discuss the contradictory thoughts in the literature regarding this topic.
Thompson (2006) using a sample of over 27,000 citing-cited US patent pairs investigated if the new patent information is geographically bounded. Evidence was found for the statement that knowledge spillovers are geographically localized in the US based on the patent data. Jaffe et al. (1993) using US patent citations data of 1975 and 1980 also found that knowledge spillovers are geographically bounded, in particular for the 1980 patent data. They established that it is more likely that inventors cite other inventors who are in their spatial proximity. Besides the use of patent data, many authors have used cluster analysis to explain the importance of geographical proximity. Being part of a cluster allows firms to operate more productively in
different ways: sourcing inputs; accessing information, technology and needed institutions; coordinating with related companies and measuring and motivating improvements (M. E. Porter, 1998). Empirical evidence is for example provided by Gilbert, McDougall, and Audretsch (2008) by using a sample of 127 independent U.S. new ventures founded between 1990 and 2000. They discovered that industry clustering has a significant and positive impact on product innovation with a level of significance at p<.05.
The underlying idea behind the findings that knowledge spillovers are limited to spatial proximity is that knowledge is partially tacit. The transfer of tacit knowledge requires frequent face-to-face interaction (Hippel, 1994). Corresponding empirical studies also found that spatial proximity between cooperation partners is a necessary requirement for the spillover of knowledge (Alcácer & Chung, 2007; Audretsch & Feldman, 2004; Döring & Schnellenbach, 2006; Fernandes & Ferreira, 2013; Fischer & Varga, 2003; Jaffe et al., 1993; Phene & Tallman, 2014). Furthermore, these findings apply to countries other than the United States. For example, Fischer and Varga (2003) find evidence for Austria.
However, other authors have argued that the influence of spatial proximity on knowledge exchange might be exaggerated (Boschma, 2005; Gertler, 2003; Weterings & Boschma, 2009). Weterings and Boschma (2009) answered three research questions, including the following question “is the positive effect of face-to-face interactions on the innovative performance of firms strengthened by spatial proximity between the organisations?” To answer this question, they gathered cross-sectional data by a survey among 265 Dutch software firms and they found that spatial proximity results in more interactions, but it does not strengthen the effect of face-to-face interactions on the innovation performance of software firms. Regarding this research, a limitation of this finding is that it is focused on software firms who generally provide services while this research focuses on manufacturing firms. However, Beugelsdijk and Cornet (2001) used a unique dataset from Statistics Netherlands (CBS) consisting of 1510 Dutch non-service firms. The dataset included information about innovation from the 1996 CIS2 survey and information about postal codes from the Firm Administration Register. The goal of this research was to find empirical evidence for the importance of spatial proximity between cooperating firms in the Netherlands. They were unable to find support for the hypothesis that innovative
Besides empirical findings from the Netherlands that questions the importance of spatial proximity, technological developments give new opportunities for face-to-face interaction. An example of such a technological development is the use of virtual teams (Kozlowski et al., 2012), which are groups of workers who are geographically dispersed with opportunities to have face-to-face contact through the use of electronic tools. More empirical evidence for the finding that technological innovations can reduce the importance of spatial proximity is found by Paunov and Rollo (2016), using 50,013 firm observations covering 117 developing and emerging countries over the period 2006-2011 with data from the second wave of the World Bank Enterprise Surveys (WBES). They claimed that ICT has reduced the barriers for transmitting knowledge by arguing that videoconference opportunities imitate face-to-face interactions better than any other means of communicating over geographic distances. Furthermore, Paunov and Rollo (2016) also found that industries’ internet use has a positive impact on firm performance and that internet-enabled knowledge spillovers contributes to increasing the group of innovative companies.
To conclude, many studies have acknowledged the importance of spatial proximity between cooperation partners for the spillover of knowledge. However, there is empirical evidence that knowledge spillovers are not geographically bounded anymore through the increasing opportunities to mimic face-to-face contact due to technological developments. Furthermore, there is empirical evidence that spatial proximity does not matter in the Netherlands. The EMS data that will be used in this thesis is only based on Dutch firms. This together results in the following hypothesis (H2)
Hypothesis 2: Spatial proximity between cooperating firms has no significant effect on the relationship between the number of non-R&D-cooperation fields a firm is cooperating in and product innovation.
2.4.3 The moderating effect of absorptive capacity
Cohen and Levinthal (1990) have argued that firms have to build an absorptive capacity to effectively use external knowledge in the development of new products and services. Therefore, it is expected that firms with a high absorptive capacity have a higher tendency to benefit from knowledge generated from external cooperation partners (Beise & Stahl, 1999).
Aldieri et al. (2018) found empirical evidence for the importance of absorptive capacity, using data from the EU R&D investments scoreboard between 2002 and 2010 made of 879 manufacturing firms located all over the world. They found that absorptive capacity determines the characteristics of external knowledge that firms have entry to. Clausen (2013) used data from the Norwegian and Swedish implementation of the third Community Innovation Survey (CIS 3) and found that if firms invest in the absorptive capacity it will enable them to use more external knowledge in the innovation process. In the research of Clausen (2013), the absorptive capacity is determined by the internal R&D investments, internal training and educational level of the workforce. Another research about absorptive capacity is conducted by Escribano et al. (2009) using a sample of 2265 Spanish firms from the CIS of 2000 and 2002, composed by the Spanish National Statistics Institute (INE). They found that external knowledge flows have a positive impact on the innovation performance of a firm and the magnitude of this effect depends heavily on a firm’s absorptive capacity. The independent variable absorptive capacity x external knowledge flows on the dependent variable innovation was positive (14.19) and highly significant.
In addition, Giuri and Mariani (2013) used data from the PatVal-EU Survey provided by the inventors of 6,051 European patents to explore whether educational level contributes to an individual’s capability to scout for and absorb knowledge spillovers. They found that a high educational level contributes to the absorptive capacity of individuals because it gives people the ability to admit the value of, assimilate and exploit external knowledge.
To conclude, the current literature states that the absorptive capacity of a focal firm has a positive influence on the extent of external knowledge firms have access to. This results in the following hypothesis (H3):
Hypothesis 3: the absorptive capacity of the focal firm has a positive significant effect on the relationship between the number of non-R&D-cooperation fields a firm is cooperating in and product innovation
expects that spatial proximity between a focal firm and a cooperation partner has no significant effect on the relationship between the number of non-R&D cooperation fields a firm is cooperating in and product innovation. Finally, the third hypothesis expects that the absorptive capacity of a focal firm has a significant positive effect on the relationship between the number of non-R&D-cooperation fields a firm is cooperating in and product innovation.
Chapter 3 – Methodology 3.1 Introduction
In this chapter, the methodology that is used in this research is explained. First the research method is explained. After that, the operationalization of the quantitative and qualitative analysis is discussed in a detailed way. At the end of the chapter, the research ethics are discussed.
3.2 Research method
The research method that has been used during this research is a mixed-method approach. For this approach is chosen because the qualitative analysis could give more in-depth explanations for the findings in the quantitative analysis. Regarding the quantitative method, data from the European Manufacturing Survey 2015 is used. The qualitative method consists of five interviews with respondents of five different firms. The next sections of this chapter provide more information about both the quantitative and qualitative methods that are used.
3.3 Quantitative analysis
As discussed in section 3.2, for the quantitative analysis data from the European Manufacturing Survey of 2015 has been used. The most recent version of the EMS is 2018. Unfortunately, the data of 2018 was not available at the beginning of this research. The aim of the EMS is to map the innovation performance of the manufacturing industry for different countries in Europe.
The complete EMS study is coordinated by the Fraunhofer Institute for Systems and Innovation Research (ISI). However, the study is sub-coordinated by different universities and research institutes across Europe. In the Netherlands, the Radboud University Nijmegen participates in this study. The EMS team of the Netherlands consists of Dr Paul Ligthart, Dr Peter Vaessen and Dr. Robert Kok. In this thesis, only EMS data from the Netherlands is used. Therefore, this thesis is focused on Dutch manufacturing firms only. All offices of Dutch manufacturing firms with at least 10 employees were contacted. The unit of analysis of this dataset are the business establishments of Dutch Manufacturing firms. More specific, the observation units of the EMS are the following: office managers, R&D managers and production managers of Dutch
3.3.1 Operationalisation of the quantitative analysis
In the quantitative part of this research, the different hypotheses discussed in chapter 2 are empirically tested with the EMS database. The different variables that are used to test the hypotheses are operationalized. The operationalisation table is displayed in table 1. The 7 categories (7 types of industries) used for the control variable industry can be found in table 3 in chapter 4.2.
Table 1 Operationalisation table
As shown in table 1, several questions of the EMS are used for the quantitative analysis. The different questions are shown on the next page, with the part of it that is used circled.
Variable type
Variable name Item (short description of the survey question)
min max Measurement level Question number in the EMS Dependent Product innovation New products introduced between 2012 and 2015 0 1 Nominal 9.1
Independent Number of
non-R&D-cooperation fields
Cooperation between a focal firm and other firms
0 4 Ratio 6.1
Moderator Spatial proximity Location of the partners 1 12 Ratio 6.1 Absorptive capacity Percentage highly educated employees 0 100 Ratio 15.1
Control Firm size Number of employees
0 + ∞ Ratio 21
Industry Industry the firm is operating in (7 categories) 0 7 Nominal 1.2 Percentage R&D employees Percentage of R&D employees 0 100 Ratio 15.2 Number of R&D-cooperation fields Cooperation in R&D fields 0 2 Ratio 6.1
Question 1.2 (variable industry)
Question 6.1 (variables number of non-R&D-cooperation fields, spatial proximity and number of R&D-cooperation fields)
Question 9.1 (variable product innovation)
Question 15.2 (variable percentage of R&D employees)
Question 21 (variable firm size)
3.3.2 Method for analysing the quantitative data
The data analysis method that is used is a logistic regression analysis. A logistic regression analysis can be used to describe the relationship between a set of metrically scaled independent variables and a non-metrically scaled dependent variable (Frost, 2017). Furthermore, this analysis is appropriate to analyse the effect of moderating variables (Hair, Black, Babin, & Anderson, 2013). The logistic regression analysis is conducted with the program SPSS Statistics. There are two types of logistic regression: multinomial and binary. However, if the dependent variable is of nominal measurement level, a binary logistic regression analysis should be conducted (Grace-Martin, 2010). Therefore, in this research the binary logistic regression method is used.
3.3.3 Validity and reliability of the quantitative analysis
To increase the validity and reliability of the EMS, several measures have been taken. Regarding the internal validity, first the questions are asked in a detailed way. Furthermore, trial surveys were conducted and meetings with representatives of 15 countries were organized to discuss intensively about the formulation of the survey questions and the drawing up of the questionnaire. Regarding the external validity, the EMS has offered a benchmark report at no charge where firms were able to compare themselves with other firms. Besides that, two reminders were sent to all firms to fill in the survey. Finally, regarding the reliability, no opinion questions were asked. Only questions about objective data were asked, for example about practices, facts, investments and performance figures. Moreover, Cronbach’s alpha has been used to measure the reliability of the quantitative constructs.
3.4 Qualitative analysis
The qualitative part of this research consists of five interviews with employees of five different companies. These interviews were conducted after the quantitative analysis with the aim to find more in-depth explanations for the quantitative findings. All respondents were found by using the researcher’s own network. The unit of analysis of the qualitative analysis was people with a managerial function in Dutch manufacturing firms. This unit of analysis was chosen because it is assumed that people with a managerial function have a good view about what is happening within their organisation. Unfortunately, due to the corona virus four out of five interviews have been conducted by phone. Only one interview has been held on location.
Before the interviews were conducted, the respondents were contacted. According to Bleijenbergh (2015) the best way to gain access to organisations for conducting interviews is through a personal contact form in combination with a written substantiation of the request. Therefore, in this research first the respondents were approached by phone and after that an e-mail was sent to the respondents. The e-e-mail contained information about the research and it was stressed that the interview would remain anonymous. Bleijenbergh (2015) distinguishes between two categories of interviews: semi-structured and unstructured. In this research the semi-structured approach has been chosen. This means that the questions were formulated beforehand. A benefit of this approach is that the researcher can determine which topics are
3.4.1 Operationalisation of the qualitative analysis
Just as the quantitative part of this research, also the qualitative part needs to be operationalised. First it is important to mention that this research is deductive because it is theory driven what is shown by the hypotheses in the previous chapter which are defined on the base of an extended literature review. According to Bleijenbergh (2015) a valid way to operationalise a deductive qualitative study is with tree structures. A tree structure consists of a visualisation of the concept, its dimensions and indicators appropriate to measure the concept. Four tree structures were constructed, which are displayed in appendix II.
3.4.2 Methods for analysing the qualitative data
As method for analysing the interviews, first the interviews were transcribed literally. This is in line with Bleijenbergh (2015) because she argues that you have to write down your received information as detailed as possible to fully benefit from the obtained information. After transcribing, the interviews were encoded with the help of the tree structures. This has made the interpretation of the information obtained in the interviews easier because it gave a structured overview of the differences and similarities in the answers given by the respondents.
3.4.3 Validity and reliability of the qualitative analysis
Several actions have been taken to optimise the reliability and validity of the qualitative analysis. First, a benefit of semi-structured interviews is that all participants are asked the same questions, what has a positive impact on the reliability (Bleijenbergh, 2015). Furthermore, to increase the external validity participants working in different industries were selected to control for industrial differences. In addition, the topics discussed during the interviews were based on an extensive literature review. This increases the construct validity of the qualitative data. Finally, the transcripts were sent back to the interviewees to avoid misunderstandings in the interpretations of certain statements. One respondent has changed the transcript a little bit because it contained misinterpretations. According to Bleijenbergh (2015), this also increases the reliability of the qualitative analysis.
3.5 Research ethics
When conducting the research, the researcher has made sure that he was acting ethically. First, regarding the interviews several ethical measures were taken. The participants had the freedom to withdrawn from the research at any time. Also, all participants were informed about the purpose, methods and possible outcomes of the research at the beginning of the interviews. In
addition, the participants were asked in advance whether they agreed to the recording of the interviews and the interviews have been made anonymous in this report.
Zooming in on the quantitative part, first the respondents are anonymous to the researcher. Also, participation in the EMS was voluntary. Unfortunately, the participants of the EMS were not informed about the purpose of this particular research because the survey is not conducted by the researcher himself. Therefore, to act as ethical as possible regarding this issue, the researcher has only received the variables that were demonstrated in the operationalisation table in table 1. Regarding the overall research, first all chapters were reviewed multiple times in cooperation with the supervisor to ensure integrity. On the base of the feedback provided by the supervisor adjustments were made. Also, the researcher has prevented plagiarism through the use of correct references. In addition, the data analysis has been carried out entirely by the researcher himself. Moreover, this research is not sponsored by an external party. This has prevented that findings that could be harmful to particular parties were manipulated. Therefore, the aim to get theoretical and practical results in a fair way has been retained. A signed research integrity form can be found in appendix X.
Chapter 4 – Data analysis 4.1 Introduction
Chapter 4 is dedicated to the data analysis. It starts with the quantitative analysis with describing the sample characteristics. Then, the variables are constructed. After that, the univariate analysis is demonstrated. It gives an overview of the descriptive statistics of each individual variable. Subsequently, the bivariate analysis is conducted and finally the multivariate analysis is done. The multivariate analysis consists amongst others of the hypotheses testing. After the quantitative analysis, the qualitative analysis in demonstrated. The chapter ends with a combined conclusion of both analyses.
4.2 Sample characteristics
The EMS 2015 dataset consists of 177 respondents (N=177). First, the size of the different firm’s participating in the survey is analysed. As demonstrated in table 2, the mode of the variable firm size is 20 to 49 employees. More precise, 41,8% of the total participating firms has between 20 and 49 employees. Furthermore, 62.7% of the firms has less than 50 employees. Therefore, most of the participating firms are small firms. In addition, only four multinationals (>250 employees) were participating in the survey. This is in line with the representation of firms in the Netherlands, because 99% of the Dutch firms has less than 250 employees (MKBservicedesk, 2020).
Frequency Percent Valid percent Cumulative percent Less than 20 employees 37 20.9% 20.9% 20.9% 20 to 49 employees 74 41.8% 41.8% 62.7% 50 to 99 employees 43 24.3% 24.3% 87.0% 100 to 249 employees 19 10.7% 10.7% 97.7% 250 or more employees 4 2.3% 2.3% 100.0% Total 177 100.0 100.0
Table 2 Firm size of companies participating in the EMS
However, as discussed in section 3.3, only firms with at least 10 employees were contacted to participate in the survey. Therefore, it can be expected that only firms with 10 or more
employees are participating in the EMS. To test this, another analysis is conducted which measures the number of employees per firm exactly. This analysis is done with the item “number of employees in 2014.” In line with the expectations, all participating firms have 10 or more employees. The most common number of employees is 20. To be precise: 8 of the 177 participating firms has 20 employees.
In addition to the number of employees of the participating firms, it is analysed in which industries they are active. Table 3 gives an overview of the industries the participating firms are working in. However, 2 missing items were found in the dataset, what means that two respondents did not fill in the question about type of industry. Because only manufacturing firms were approached for participating in the EMS, it is assumed that the two firms that did not fill in this question are also manufacturing firms. Therefore, these missing values are not deleted from the dataset.
Industry Frequency Percentage
Metal and metal products 37 20.9%
Food, beverages and tobacco 18 10.2%
Textile, leather, paper and board 22 12.4%
Construction, furniture 13 7.3%
Chemical (energy and non-energy) 22 12.4%
Machinery, equipment transport 31 17.5%
Electronic and optical equipment 32 18.1%
Total 175 98.9%
System missing 2 1.1%
Total 177 100.0%
Table 3 Overview type of industry
As shown in table 3, metal is the best represented industry in the EMS 2015 dataset, followed by the electronic and the machinery industry. However, according to CBS (2018), the chemical, electronic and machinery industry were the biggest industries in the manufacturing branch in
conclusion that the representation of the different industries in the survey is not completely in line with the representation of the different industries in the Dutch manufacturing branch.
4.3 Variable construction
This section is an extension on the operationalisation table made in section 3.3.1 because it describes the construction of the variables. The same sequence as in the operationalization table is used, what means that the construction of the dependent variable is described first.
4.3.1 Construction of the dependent variable
The dependent variable of this research is product innovation. Product innovation is measured in the EMS with question 9.1. This question measures if a company has introduced products that were new to the company between 2012 and 2015. The answer options on this question are only ‘yes’ and ‘no.’ Therefore, the dependent variable of this research is a dichotomous variable. For a dichotomous variable, there is no possibility to check the consistency between the different items. Thus, a reliability analysis is not required.
4.3.2 Construction of the independent variable
The independent variable of this research is the number of non-R&D-cooperation fields a firm is cooperating in. Question 6.1 of the EMS is used for constructing this variable. This question measures six types of cooperation fields. However, two types are about cooperation in R&D while the focus is on non-R&D-cooperation. Therefore, these two types are not considered by constructing this variable. As a consequence, four items remain by measuring the number of non-R&D-cooperation fields: cooperation in purchasing, cooperation in production, cooperation in sales/distribution and cooperation in service. All these four items have a min of zero (no non-R&D-cooperation in a particular field) and a max of 1 (non-R&D-cooperation in a particular field).
These four items must be merged into one variable to construct the independent variable number of non-R&D-cooperation fields. Before adding the items together, a scale analysis is conducted to check the internal consistency between these items. The scale analysis gives a Cronbach’s alpha of .613. Table 4 shows the Cronbach’s Alpha if items are deleted.
Table 4 Cronbach’s Alpha if item deleted non-R&D-cooperation
As shown in table 4, it is impossible to increase the Cronbach’s Alpha by deleting an item. Thus, the Cronbach’s Alpha of this variable remains .613. As a consequence, the constructed variable number of non-R&D-cooperation fields consist of the 4 types as shown in table 4. This value is acceptable because according to Hair et al. (2013) values between .6 and .7 are considered as the lower limit of acceptability.
4.3.3 Construction of the moderating variables
Spatial Proximity
Spatial proximity is measured with question 6.1. If a firm is cooperating in one of the four fields as discussed by the construction of the independent variable, the subsequent question is the location of the cooperation partner, divided into three subcategories: regional (<50km), national (>50km) and abroad. For the subcategories, three sub variables are constructed. To construct the sub variables, first the total number of cooperation for each subcategory is counted. After that, the total number of cooperation for each subcategory is divided by the total number of cooperation for all subcategories together. This has resulted in three sub variables which are used for measuring the construct spatial proximity: relative regional cooperation, relative national cooperation and relative abroad cooperation.
Absorptive capacity
Absorptive capacity is measured by the percentage of highly educated employees (question 15.1). As this question has only one item, no scale analysis is needed.
4.3.4 Construction of the control variables
Item Cronbach’s Alpha if item deleted
Purchasing cooperation .586
Production cooperation .506
Sales/distribution cooperation .482
Business sector
For measuring the control variable business sector, a categorical variable is created which divide the types of industry into 7 categories. The different categories are shown in table 3.
Number of R&D-cooperation fields
The control variable number of R&D cooperation fields is measured with two items of question 6.1. These two items are: “cooperation in research and development with buyers or suppliers” and “cooperation in research and development with research institutes.” These two items are added together to construct the control variable number of R&D-cooperation fields.
Percentage of R&D employees
The control variable percentage of R&D employees is measured with question 15.2. This question asks for the percentage of employees working in the area of research and development.
4.4 Univariate analysis
This section discusses the descriptive statistics of the variables used in this research, amongst others by checking for normality of the metric variables. In line with Field (2018), the critical values for normality of the metric variables are a skewness and kurtosis between -3 and +3.
Dependent variable
Because the dependent variable product innovation is a dichotomous variable, only the frequencies of this variable are discussed. As shown in table 5, 61.6% of the firms participating in the survey have conducted product innovations between 2012 and 2015.
Table 5 Frequencies of product innovation (dependent variable)
Independent and moderating variables
The independent variable (number of non-R&D-cooperation fields) is a metric variable, just as the moderating variable spatial proximity, measured by 3 sub variables and the moderating
Product innovation Frequency Valid percent
0 – no 68 38.4 %
1 – yes 109 61.6 %
variable absorptive capacity, measured by the percentage of highly educated employees. Table 6 displays the descriptive statistics of these variables.
Table 6 Descriptive statistics of metric variables
By checking for normality, it can be concluded that almost all variables are normally distributed because they have a skewness and kurtosis between -3 and +3. The only exception is the variable percentage of highly educated employees because this variable has a kurtosis of 4.604. Therefore, this variable needs to be transformed. Several transformations were made. The output of the different transformations is shown in appendix III. The square root transformation gave the best values of skewness and kurtosis. Therefore, by measuring absorptive capacity the variable SQRT_highly_educated will be used. Table 7 shows the skewness and kurtosis of this new variable.
Table 7 Old and new skewness and kurtosis variable highly educated employees
Variable Mean Median Mode Sd Min Max Skewness Kurtosis Number of non-R&D-cooperation fields 1.5028 1.0 .0 1.31479 0 4 .358 -1.094 Relative regional cooperation .4959 .5000 .00 .43883 0 1 .027 -1.741 Relative national cooperation .5130 .5000 1.00 .44252 0 1 -.041 -1.766 Relative abroad cooperation .4360 .3333 .00 .44001 0 1 .229 -1.721 Percentage of highly educated employees 16.0339 10.00 10.00 14.66439 0 80 2.025 4.604
Variable Skewness Kurtosis
% highly educated employees (“old” variable) 2.025 4.604
Control variables
Two of the control variables are metrically scaled: number of R&D-cooperation fields and percentage of R&D-employees. The descriptive statistics of these control variables can be found in table 8. Based on the table it can be concluded that both variables are normally distributed because their skewness and kurtosis are between -3 and +3. The mean of the variable percentage of R&D-employees is 5.58, what means that the firms participating in the EMS have on average 5.58% of the employees working in the area of Research & Development.
Table 8 Descriptive statistics of metric control variables
The other two control variables (type of industry and firm size) are non-metrically scaled variables. Therefore, only their frequencies are meaningful in the univariate analysis. The frequencies of both variables are already discussed in section 4.2.
4.5 Bivariate analysis
In the previous section, all variables have been analysed independently. Thereby the normality of the variables is analysed, and a transformation has been made for one variable that was not normally distributed. This section will test whether and how correlations exists between the different variables by monitoring for multicollinearity. It is suggested that multicollinearity exists when variables are correlated with R-values above .80 or .90 (Field, 2018). A Pearson correlation matrix is conducted to check for multicollinearity. The Pearson correlation matrix with the R-values included can be found on the next page, in table 9.
Variable Mean Median Mode Sd Min Max Skewness Kurtosis Amount of R&D-cooperation .9096 1.0 .0 .80677 0 2 .166 -1.444 Percentage of R&D-employees 5.5847 5.000 .00 5.81887 0 25 1.238 .935
* p < .05, ** p < .01 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1. Product innovation 1 .126 - .122 .158 .217* .022 .344** .198** .182* - .075 .038 -.051 .002 .055 .094 -.047 2. Number of Non-R&D-cooperation fields 1 -.176 .213* .231* .144 .263** .066 .243** -.162* -.055 .042 .110 .002 .203** -.098 3. Rel_reg_c 1 -.218 -.348** -.057 .005 -.008 .211* .093 .036 .070 .025 -.025 -.165 -.005 4. Rel_nat_c 1 -.002 -.077 .246** -.028 .045 -.026 .001 .044 -.005 -.064 .062 -.022 5. Rel_abr_c 1 .231* .217* .222* .154 -.244** -.098 .075 .095 -.009 .171 .001 6. SQRT_highly_edu 1 .228** .291** -.075 -.199** -.037 -.065 -.065 -.046 .292** .089 7.Amount of R&D-cooperation 1 .267** .248** -.105 .034 -.005 .055 .038 .065 -.045 8. Percentage R&D-employees 1 -.042 -.074 -.068 -.049 -.044 .008 .004 .193* 9. Firm size 1 -.110 .042 -.035 .084 .068 .106 -.107 10. Metal 1 -.175* -.196** -.147 -.196** -.240** -.245** 11. Food 1 -.128 -.096 -.128 -.157* -.160* 12. Textile 1 -.107 -.144 -.176* -.179* 13. Construction 1 -.107 -.131 -.134 14. Chemical 1 -.176* -.179* 15. Machinery 1 -.219** 16. Electronic 1
All variables have low values of collinearity, what is unavoidable according to Field (2018). Furthermore, low levels of collinearity are little threatening to the model estimates. The highest correlation between variables is -.348. This value is much lower than the value for multicollinearity of .80. Therefore, based on the R-values of Pearson’s correlation matrix it can be concluded that no multicollinearity exists between the different variables.
Besides checking for multicollinearity, the bivariate correlation table is also appropriate to check whether the expected relationships already can be found in the analysis. To determine the effects, the ranges suggested by Field (2018) are taken into account: tiny (values between +/- 0 and 0.1), small to medium (values between +/- 0.1 and 0.3), medium to large (values between +/- 0.3 and 0.5) and large (values greater than +/- 0.5).
The first hypothesis expects a positive relationship between the number of non-R&D-cooperation fields a firm is cooperating in and product innovation. Table 9 displays a small to medium positive correlation (.126) between these two variables. However, this correlation is not significant. The second hypothesis expects that spatial proximity between firms does not matter anymore due to technological developments. The correlation table gives a significant small to medium negative correlation of regional cooperation (-.122), a non-significant small to medium positive correlation of national cooperation (.158) and a non- significant small to medium positive correlation of abroad cooperation (.217) on product innovation. This is in line with the hypothesis that spatial proximity between firms has no significant effect on the relationship between non-R&D-cooperation and product innovation. Finally, the last hypothesis expects that a firm’s absorptive capacity, measured by the percentage of highly educated employees, has a positive impact on the relationship between non-R&D-cooperation and product innovation. The correlation table indicates an insignificant tiny positive correlation between the percentage of highly educated employees and product innovation with a value of .022. In conclusion, all expected directions of the hypotheses are confirmed in the correlation table, but they are all non-significant. Therefore, based on the bivariate analysis only hypothesis 2 can be confirmed.
After checking for the correlations of the hypotheses, the correlations of the control variables are analysed. The control variable number of R&D-cooperation fields has some significant correlations. First, it is significant positive medium to large correlated (.344) with product innovation. Furthermore, it is also positive and significant small to medium correlated (.263)
with the independent variable, what indicates that most of the firms that are cooperating in R&D, also cooperate in non-R&D. This makes the importance of R&D-cooperation as control variable clear, because it prevents that a false relationship between non-R&D-cooperation and product innovation will be found in the multivariate analysis. The second control variable, percentage of R&D-employees is also significant and positive small to medium correlated (.198) with the dependent variable, what indicates that the percentage of R&D-employees also has a significant effect on product innovation. Moreover, both the variables number of R&D-cooperation fields and percentage of R&D-employees are positive and significant small to medium correlated with the variable percentage of highly educated employees. In addition, the control variable firm size has a significant positive small to medium correlation (.182) with product innovation. This indicates that the bigger the firm, the more new products are introduced. The reason for this could be that big firms have more resources to introduce new products. Firm size also has a significant positive small to medium correlation (.243) with non-R&D cooperation, what indicates that the bigger the firm, the more they cooperate in the fields of purchasing, production, sales/distribution and service. Moreover, abroad cooperation has a significant positive small to medium correlation with the variable percentage of highly educated employees (.231). This seems logic because when companies have more capabilities, they are more able to transfer knowledge across long distances (Giuri & Mariani, 2013).
Finally, the noteworthy correlations of the control variable industry are discussed. The metal industry has a significant negative small to medium correlation (-.162) with non-R&D-cooperation, what indicates that in the metal industry non-R&D-cooperation is less common. An opposing correlation occurs between the machinery industry and non-R&D-cooperation because there is a significant positive small to medium correlation (.203), what indicates that non-R&D-cooperation is more common in the machinery industry. Moreover, there exists a significant positive small to medium correlation (.292) between the machinery industry and the variable percentage of highly educated employees, what indicates that in the machinery industry the number of highly educated employees is high.
4.6 Multivariate analysis
