Strategy: an Employee- level Perspective” “The Unintended Consequences of Secrecy as a Knowledge Protection

(1)

Master Thesis

MSc BA – Strategic Innovation Management

Quantitative empirical study on the impact of secrecy on employee knowledge sharing

“The Unintended Consequences of Secrecy as a Knowledge Protection

Strategy: an Employee-level Perspective”

Sebastiaan Willem Rutger Dekker

University of Groningen

S2497700

June 24th_{, 2019}

Supervisor: Pedro de Faria Co-assessor: Philip Steinberg

(2)

2

Abstract

To effectively capture value from innovations, firms use secrecy as a tool for protecting innovation-specific knowledge that provides a competitive advantage. This makes secrecy especially relevant for large firms operating in R&D intensive industries, where knowledge is a key resource. Strategy literature acknowledges the importance and effectiveness of secrecy as a strategic knowledge protection instrument on the firm-level, but disregards its micro-foundations. An unintended side-effect of secrecy resides in a reduction of knowledge sharing between employees, due to regulations and complexities that accompany its implementation. Because knowledge sharing is essential for innovation, this hampers the innovation capabilities of firms. I examine the effects of secrecy on employee knowledge sharing through a survey conducted at 2 firms operating in the agricultural enhancement industry in the Netherlands. Hypotheses were tested on a sample of 55 employees active in innovation. The results partially support the proposed relationship by indicating a negative effect of secrecy restrictions on employee knowledge sharing within and outside teams. However, for other dimensions of secrecy and knowledge sharing the relationship is positive or non-significant. The findings contribute to the theoretical and managerial understanding of secrecy as a knowledge protection instrument, and encourages future studies to acknowledge the unintended effects of secrecy on knowledge sharing at the employee-level.

Keywords: Secrecy; Knowledge sharing; Innovation; Knowledge Protection; Secrecy Management; Appropriation

(3)

3

1. Introduction

In the field of strategy and management research, the importance of innovation is consistently recognized as a critical antecedent of superior firm performance (Calantone, Cavusgil, & Zhao, 2002; Han, Kim, & Srivastava, 1998; Jansen, Van Den Bosch, & Volberda, 2006). Firms that continuously create valuable new products or processes through intensive research and development (R&D), and are able to successfully commercialize those inventions, have an opportunity to outperform competitors (Ceccagnoli, 2009). However, merely creating value through innovations is insufficient to realise a sustained competitive advantage. Competitors could engage in imitative behaviour that reduces profits for the focal firm (Ceccagnoli, 2009; James, 2014; Teece, 1986), whereby firms would have low incentive to innovate (Levin et al., 1987). Therefore, it is essential for firms to effectively appropriate value from innovation-specific knowledge, by adopting knowledge protection mechanisms (Bos, Broekhuizen, & De Faria, 2015). These mechanisms are aimed at complicating and discouraging imitation by rivals or new entrants, to increase the firm’s value appropriation and innovation performance, thereby potentially upholding a competitive advantage (James, Leiblein, & Lu, 2013).

Generally, strategy literature recognizes two groups of knowledge protection mechanisms formal and informal knowledge protection mechanisms. Formal mechanisms include patents, copyright and trademarks, which offer legal protection against imitation. As primary formal mechanism, patents are tangible instruments that protect the innovation’s value through temporary, institutionally granted, innovation-specific monopolies for firms to commercialize on. However, in exchange the firm provides public disclosure of the information regarding the patented technology which may benefit competitors (James, 2014; Levin et al., 1987). Additionally, their effectiveness mainly depends on the strength of the institutional regime (Teece, 1986) and the ability of competitors to legally circumvent patented technologies in imitation efforts (Levin et al, 1987). Other obstacles are the considerable costs associated with the application process, annual renewal and legal enforcement of potential patent infringements (Hussinger, 2006).

(5)

5

loss of competitive advantage (Bos et al., 2015; Hannah, 2005). Therefore, Liebeskind (1997) and Bos et al. (2015) emphasize that on the long term secrecy does entail costs in strategic assessment and management processes to prevent these unintended spillovers. Other disadvantages or costs of secrecy remain ambiguous. The implementation of secrecy is increasingly relevant, since rapidly changing technologies enable competitors to more efficiently circumvent or ‘invent around’ patents, which may deteriorate their effectiveness (Arundel, 2001; Cohen, Nelson, & Walsch, 2000; Levin et al, 1987). Thereby, the use of secrecy as an alternate protection mechanism is of substantial importance for R&D-active, technology-driven firms that rely on the creation and exploitation of knowledge to create value through innovation (Bos et al., 2015). Having protected firm-specific knowledge is a key determinant of establishing and sustaining a competitive advantage (Schmidt, 2005). However, the effectiveness of secrecy, patenting and other knowledge protection instruments is highly dependable on environmental, firm-specific and industry characteristics (James et al., 2013; Schilling, 2010; Teece, 1986).

Considering the prevalence of secrecy across all industries (Cohen et al., 2000), it is remarkable that the concept remains relatively undeveloped when compared to the comprehensive research streams focused at patenting (James et al., 2013; Sofka, de Faria, & Shehu, 2018). An explanation for this unbalance is found in the wide availability of codified patent information, which has been registered at a patent office and is tangible in nature. Metrics of secrecy, however, are often highly tacit, intentionally undisclosed or unavailable. Therefore it is relatively less straightforward to quantifiably measure secrecy in contrast to more explicit forms of knowledge protection, which results in the obscurity of the concept. The niche of research that does address secrecy, is primarily aimed at understanding how secrecy as a knowledge protection mechanism can be strategically utilized to effectively capture value from innovations (e.g. Arundel, 2001; Bos et al., 2015; James et al., 2013; Hussinger, 2006; Katila et al., 2008; Sofka et al., 2018). Strategic decision-making on protection mechanisms by innovation-focused top managers in large multinational firms are identified as the drivers of innovative success (Hult, Hurley, & Knight, 2004; Klein & Sorra, 1996; Nemeth, 1997).

(6)

6

restrict knowledge flows, and exchanging knowledge within the firm for innovation purposes. The goal of secrecy is to prevent external spillovers, however in applying regulations to enforce secrecy, these strategic measures may unintendedly restrict internal spillovers between employees, which hampers innovation. More specifically, a reduction in knowledge sharing emerges from the increasing pressure to follow rules and regulations that create knowledge asymmetries. Employees could generally face difficulties in interpreting how and with whom they are allowed to share knowledge according to the rules and regulations, and therefore are not able to effectively share knowledge. Subsequently, when it is clear with whom and how to share knowledge, it is up to personal judgment if employees trust specific co-workers to conscientious handle sensitive knowledge. Additionally, when asymmetries in knowledge access create distrust or negative tensions in the workplace, secrecy measures constrain employees in sharing knowledge (Isaksen & Ekvall, 2010).

Regarding the knowledge sharing process, whereas new technologies create opportunities to more rapidly and effectively share or manage knowledge (Malhotra, 2000; Santoro, Vrontis, Thrassou, & Dezi, 2017; Thierer, 2014), it becomes increasingly complex to safely share knowledge through protected channels. Therefore, sharing sensitive knowledge or secrets is more difficult for employees if they have to account for the risk of unintended leakage, which increases transaction costs in terms of effort and decreases the willingness to share. Moreover, secrecy measures include physical constraints in the workplace, which reduces the flexibility of employees in sharing knowledge through preferred methods (Sofka et al., 2018). These effects generate barriers to knowledge sharing which potentially obstruct employees in reaching their full innovative potential through internal knowledge sharing.

Knowledge sharing of both explicit and tacit knowledge of employees within an organization has structurally been identified as an imperative to the firm’s innovativeness, performance, competitiveness and market position (Riege, 2005; Wang & Wang, 2012). Therefore, an unintended decrease in employee knowledge flows due to secrecy measures negatively affects employee innovativeness. I argue that secrecy policies lower the potential for created value due to reduced knowledge sharing, and thereby is paradoxically opposing the intended positive effects of effectively capturing value from innovations. The proposed effects of secrecy on employee knowledge sharing is disregarded in current academic literature. Therefore, it is unclear how employees are affected by the use of secrecy in their daily work, which is critical to investigate to fully comprehend the multiple facets of the knowledge protection mechanism.

(7)

7

comprehensive understanding of the effects of adopting secrecy as a knowledge protection mechanism. From an employee perspective, the implementation of secrecy policies entails (a) knowledge asymmetries; (b) tensions in following and interpreting regulations; (c) tensions in trust assessments; (d) increased complexity, effort and costs of knowledge sharing and (e) physical constraints in the workplace. I propose that these factors negatively influence the level of knowledge sharing between employees in firms that utilize high levels of secrecy as a knowledge protection instrument. Therefore, I formulated the main research question for this study as follows:

“What is the unintended side-effect of secrecy as a knowledge protection mechanism on employee knowledge sharing?”

In addition to internal knowledge sharing, the side-effects effects of secrecy may also refer to knowledge sharing with colleagues or professionals located outside the boundaries of the organization. The primary objective of secrecy is to reduce external information flows, which also entails a reduction in sharing non-sensitive or professional knowledge which fosters collective learning and innovation opportunities with actors beyond the boundaries of the organization (e.g. conferences, professional meetings and training sessions). When employees feel restricted to engage in knowledge sharing, firms may not exploit partnerships to their full extent. Therefore it is relevant to investigate if the effects of secrecy differ between internal or external knowledge sharing behaviour of employees. The secondary research question is stated as follows:

“How does the unintended side-effect of secrecy on employee knowledge sharing affect knowledge sharing outside the boundaries of the firm?”

(8)

8

2. Theoretical Framework

2.1. Knowledge Protection

Firms innovate to create value and outperform competitors both financially and in market share (Ceccagnoli, 2009). To sustain this advantage, it is essential to protect value-creating resources against imitation and effectively appropriate value (Ceccagnoli, 2009; James, 2014; Teece, 1986). Schilling (2010) defines appropriability as “the degree to which a firm is able to capture the rents from its innovation” (p. 188). According to resource-based view, for resources to offer a source of sustained competitive advantage they should be valuable, rare, inimitable and non-substitutable (VRIN framework; Barney, 1991). Grant (1996) extends this construct and argues that knowledge is the primary resource to sustain a competitive advantage (knowledge-based view [KBV]). Therefore, long-term strategic decision-making should focus on generating and acquiring new knowledge. Accordingly, the firm should focus on selecting and protecting its VRIN knowledge resources that create superior value and sustain the competitive advantage, and effectively appropriate value by installing effective knowledge protection mechanisms. (Ahmad, Bosua, & Scheepers, 2014). When value cannot be captured effectively, and the created value is appropriated by other actors in the horizontal or vertical value chain, firms rapidly lose incentive to invest in new knowledge creation and innovation (Levin et al., 1987). The primary goal of knowledge protection strategies is therefore to reduce the threat of imitation, by making it as complex and costly as possible, thereby enabling firms to effectively capture value. (Bos et al., 2015; Sofka et al., 2018).

(9)

9

Firms implement various mechanisms to prevent sensitive knowledge from spilling over to external parties. Two prevalent groups of knowledge protection mechanisms have emerged in practice and literature: formal and informal mechanisms. Formal protection mechanisms consist of patents, copyrights and trademarks. Patents protect a technology or invention, copyright protects the original creator of an artistic or literary work and trademarks include the protection of elements associated with a certain product or brand that differentiate it from other brands (Schilling, 2010). They offer legal protection against misuse and can be applied for at a country’s institutional system. The other group of informal mechanisms mainly refers to secrecy, lead time advantages, retention of key employees, complexity of design and access to complementary assets. Informal mechanisms create barriers to market entry or increase the costs of imitation (James et al., 2013), but often cannot be legally enforced. Regarding the scope of this research, I focus on patenting as the most widely implemented formal mechanism and secrecy as the most relevant informal mechanism

The effectiveness of protection mechanisms is highly dependable on external factors, and therefore varies greatly for the innovation’s nature, environment and characteristics. Firstly, the successful implementation of knowledge protection instruments is considerably influenced by the strength of the institutional environment of a country or region (James et al., 2013; Teece, 1986). In regimes with a relatively strong and adequate legal enforcement of intellectual property (IP) protection laws (i.e. strong appropriation regime), formal mechanisms such as patenting are more effective than in those institutional environments with a relatively weak legal enforcement of IP protection laws (i.e. weak appropriation regime). Consequently, in weak appropriation regimes costs of imitation for competitors are marginally lower, and informal mechanisms may create more effective barriers to imitation since IP protection laws are not fully enforced.

(10)

10

and possibilities for transferring knowledge (i.e. the nature of the technology) determines the choice for effective knowledge protection instruments.

A third factor that influences the effectiveness of protection mechanisms, is the complexity of the innovation (James et al., 2013). Less complex innovations, with a low number of technical elements that can be patented, have limited overlap with other firm’s patents. Consequently, within industries that produce less complex outcomes, a firm can make stronger claims on the novelty of patented technologies, and is more inclined to use formal protection mechanisms. For innovations that are highly complex and consist of a high number of patentable elements that may be acquired through cooperative agreements or licensing, the novelty of the innovation is more difficult to proof in an infringement lawsuit, which increases the probability that firms choose informal mechanisms as the dominant value protection strategy (Cohen et al., 2000).

In comparison, firms select between patenting and secrecy as the appropriate knowledge protection instrument based on their ability to effectively protect the knowledge at low cost (Bos et al., 2015), while considering innovation-specific and external factors. Furthermore, the firm would be restricted to either using secrecy or patents as a protection mechanism when the innovation has been introduced to the market, because of the disclosure requirements of patenting (Hussinger, 2006). However, this is a rather generalized view, since often the minimum disclosure requirements for patenting allow for secrecy about essential processes for the innovation. Additionally, there are opportunities to capture value on both the explicit and tacit technological elements of the innovation by applying formal and informal knowledge protection mechanisms simultaneously or sequentially (Bos et al., 2015; James et al., 2013; Sofka, Shehu, & de Faria, 2014). Moreover, Arundel (2001) argues that knowledge protection tools are not mutually exclusive; for large firms a common strategy is to apply secrecy during the development phase to prevent unwanted spillovers, and invest in patenting after the launch phase to legally sustain the competitive advantage. Whereas scholars identify this potential for synergies and complementarity of patents and secrecy, there is a lack of knowledge on how firms sequence and combine different formal and informal knowledge protection mechanisms (Bos et al., 2015). In sum, the extent to which firms engage in knowledge protection and choice of formal or informal mechanisms, mainly depends on the strategic focus of policy-makers (cooperating or protecting) and industry conditions. Ultimately, the choice between patenting and secrecy as protection mechanisms relies on the associated financial costs, conditions of the institutional environment, nature of the underlying knowledge and complexity of the innovation. 2.2. Patents

(11)

11

useful, novel and non-obvious (Schilling, 2010). By applying for a patent, the firm is granted a temporary monopoly on the commercial exploitation of a new technology or innovation (Hussinger, 2006). When firms discover that competitors utilize the patented knowledge or technology to imitate a product or process, they can issue infringement lawsuits to stop their activities and receive financial compensation for the misappropriation of value. Firms may have varying incentives to patent a specific technology: they can commercialize the invention in their products or services or license patents to other parties to generate value (Cohen et al., 2000). Moreover, they may apply for patents to prevent competitors from utilizing a similar technology or use them as leverage in negotiations (Cohen et al., 2000; Hussinger, 2006).

(12)

12

Due to the associated costs and the obligatory disclosure of innovation-specific information, firms may prefer informal mechanisms over patents in protecting key knowledge resources (Hussinger, 2006). This is especially relevant in weak appropriability regimes, where patents are more difficult to uphold (James et al., 2013; Teece, 1986). However, depending on their fit with the innovation and external conditions, patents often prove highly valuable to firms.

2.3. Secrecy

The strategic implementation of secrecy is an effective alternative to patenting and widely implemented as knowledge protection instrument (Arundel, 2001). Hussinger (2006) argues that secrecy is the preferred protection methods for all types of innovations across industries. James et al. (2013) define secrecy as “…a firm’s efforts to protect the uniqueness of an innovation by withholding its technical details from public dissemination” (p. 1126). It aims at prevent spillovers that could diminish this competitive advantage, by making imitation as expensive and unpredictable as possible (Bos et al., 2015; Hannah, 2005). The construct implies a direct impediment on knowledge sharing by limiting information flows about key knowledge (Schmidt, 2005). Hereby it prevents unintended leakages, protects the critical knowledge that sustains a competitive advantage and increases the potential for appropriating value from the innovation outcomes (James et al., 2013).

The main advantage of secrecy is that the core knowledge and technologies that provide value to the innovation remains hidden and exclusively available within the firm (De Faria & Sofka, 2010; Hussinger, 2006). Thereby, it is a seemingly inexpensive measure to reduce spillovers, which may be sustained indefinitely and includes low initial costs to apply (Bos et al., 2015; Hussinger, 2006). Consequently, small firms with limited financial resources often prefer secrecy over patenting (Arundel, 2001; Cohen et al., 2000; Thomä & Bizer, 2013). However, Liebeskind (1997) argues that the costs for designing and installing secrets, and the costs related to monitoring or enforcement throughout the organization may be severe and are often not accounted for in strategic decision-making. For instance, continuous assessment of costs and future benefits is required to effectively analyze and exploit the value of a secret on the long term. Strategic secrecy management to enable sustained protection is necessary, ever-changing and resource-intensive (Bos et al., 2015).

(13)

13

mechanisms are less effective by a lack of legal enforcement of infringement or imitation practices (Teece, 1986). However, increasing legal requirements in the disclosure of firm-specific information to shareholders could reduce the effectiveness of secrecy as an independent protection tool, if the sensitive knowledge creating the competitive advantage originates from resources or capabilities that are mandatory to disclose (Sofka et al., 2018).

(14)

14

In addition to retaining key employees, a second mechanisms to minimize the risk of unintended knowledge spillovers is to apply formal instruments that enforce secrecy. These instruments include non-disclosure agreements and non-compete clauses in employee contracts, in which individuals agree to not share information with third parties at the risk of substantial financial penalties (Hannah, 2005). Additionally, intellectual property that belongs to a firm but cannot be patented (e.g. recipes or formulas) can be formally defended as a trade secret after it has been leaked or stolen, granting the firm formal protection on its unique knowledge (Bos et al., 2015; Schilling, 2010). Similar to patenting, the effectiveness of these formal agreements largely relies on institutional environments with strong legal IP protection that facilitate these agreements (Winter, 2000).

(15)

15

3. Hypotheses Development

According to Grant (1996), knowledge is the primary resource for attaining a competitive advantage, and should be acquired and disseminated accordingly, since it is essential for sustained innovation. The most critical knowledge source within organizations is predominantly tacit and resides within employees (Bartlett & Ghoshal, 2002; Olander, 2011; Pfeffer, 1994). To foster creativity and innovation opportunities, knowledge sharing processes of employees in an organization should be encouraged (Cerinšek & Dolinšek, 2009; Pfeffer, 1994; Preiss & Spooner, 2003; Wang & Wang, 2012; Zhao, 2005). Knowledge sharing is defined as “the exchange of knowledge between and among individuals, and within and among teams, organizational units, and organizations” (Paulin & Suneson, 2012, p. 83), where one entity dissimilates and the other entity assimilates the shared knowledge through interaction (Paulin & Suneson, 2012).

When knowledge resources are utilized to innovate, it is essential for the firm to effectively capture value by installing protecting mechanisms that prevent imitation, and potentially attain or sustain a competitive advantage (Bos et al., 2015; Ceccagnoli, 2009; James, 2014; Sofka et al., 2018). Firms may decide to implement formal mechanisms, informal mechanisms, or both mechanism sequentially or simultaneously. The choice between knowledge protection mechanisms generally depends on its costs, the appropriation regime, the nature of the knowledge and the complexity of the innovation. Secrecy is a widely implemented informal knowledge protection mechanism that restricts unwanted external knowledge spillover to prevent imitation, by limiting and monitoring external knowledge flows (Schmidt, 2005). Compared to patenting it is a seemingly less expensive and more sustainable mechanism to create barriers towards imitation by managing knowledge flows. Nevertheless, a continuous risk of leakage threatens its effectiveness. Especially for firms with high levels of R&D, where newly developed information is valuable, leakage of sensitive knowledge could lead to the failure of an innovation or misappropriation of value. However, restrictions in knowledge sharing to enforce secrecy may have unobserved and unintended effects within the firm.

(16)

16

of leakage that harms the firm’s competitiveness. Hereby, secrecy creates tensions in following or violating the regulations about key knowledge in an organization that enforce secrecy, since there is varying obligation between employees to protect a secret (Hannah, 2005). Another internal mechanism that firms implement to prevent imitation is compartmentalization, which entails providing different employees with only a partial component of the sensitive knowledge (Dufresne & Offstein, 2008). This strategy provides effective protection against harmful leakages, since (former) employees that either intentionally or unintentionally disclose information to competitors do not possess the full knowledge required to successfully imitate. However, by compartmentalising information in and between departments or firms, a reduction in knowledge flows may restrict organizational learning, which reduces efficiency of employees working within R&D (Winter, 2000). These effects are lower with the use of patents than with secrecy, since it is clear what can be shared and disseminated throughout an organization since it is legally protected and creates less knowledge asymmetries.

Secondly, the legislative and operational secrecy measures also foster socially complex trust relationships. Trust may foster knowledge sharing; Dufresne and Offstein (2008) state that secrecy in an organization can generate strong organizational and interpersonal commitment of employees by establishing trust relationships through sharing valuable secrets. Trust constructs are a source of knowledge sharing and innovative behaviour among colleagues (Dirks, 1999). Employees that have been trusted with secretive knowledge can be intrinsically motivated by an increased sense of importance or empowerment (Bos et al., 2015). However, when employees distrust co-workers in carefully handling sensitive information and perceive them as a risk, they will only share it with co-workers whom are deemed trustworthy. Therefore, when rules and regulations allow an employee to share secretive knowledge within the firm, it is not a guarantee that the knowledge will be effectively shared, this often depends on the personal judgment of the employee or manager (Bouty, 2000). The individual assessment of trustworthiness creates additional knowledge asymmetries. When an employee is not included in a trust network about secretive knowledge, or does not have a clearance level to access the secret (Bos et al., 2015), this asymmetry in knowledge inspires tensions between employees and creates additional barriers to knowledge sharing. Additionally, if employees are aware of the value of certain knowledge and the lengthy measures taken to protect it, and one individual unintentionally discloses this to external parties, his co-workers may behave negatively towards him and be reluctant to share further important knowledge. Thereby this process of social retaliation, or the fear thereof, further increases pressure on employees to protect knowledge and may reduce knowledge sharing through decreasing levels of trust.

(17)

17

and efficient business processes (Malhotra, 2000; Thierer, 2014) and knowledge management systems that foster knowledge flows (Santoro et al., 2017). However, the increasing use of technology implies a challenging management problem for firms relying on knowledge as their primary source of competitive advantage. The widespread use of online communication methods, social media, cloud computing and data storage resulted in a globally accessible network of knowledge (Ahmad et al., 2014). The increase of interconnected networks of individuals and information makes protecting sensitive information from misuse or theft increasingly demanding. Accordingly, the growing importance of social media and mobile communication devices increases the rate of information dissemination. These developments increase the risk that leakages of sensitive knowledge will be rapidly identified and exploited by competitors, which complicates recovering value from unintended knowledge spillovers. To prevent leakages, sensitive knowledge can often only be transferred between individuals through protected and regulated channels. This process includes extensive monitoring of employees, and often result in extensive costs, additional constraints in the workplace, reduced employee satisfaction and a decline in employee creativity (Bos et al., 2015; Liebeskind, 1997). For instance, employees may be no longer permitted to access firm knowledge remotely due to secrecy regulations, which restricts them to work from home. Moreover, employees are often restricted in working with physical copies of sensitive information to reduce the chance of unintended leakage or spillovers (Sofka et al., 2018). Additional impediments to knowledge sharing can be found in restrictions on USB drives usage, or limited access to external websites on firm-issued devices (Hu, Dinev, Hart, & Cooke, 2012). James et al. (2013) acknowledges this effect and points out that policies that are implemented to prevent external spillovers, may reduce knowledge flows within the organization.

In sum, secrecy effectively prevents unintended knowledge spillovers by limiting external knowledge flows. However, by imposing internal rules and regulations accordingly, secrecy influences the knowledge sharing process of employees by defining how and with whom information and key knowledge can be shared. In following these regulations on sustaining secretive knowledge, employees have to be (a) concerned with whom they are allowed to share the knowledge, (b) who they trust to not misuse the knowledge and (c) the increased complexity of handling the knowledge. Knowledge sharing thereby becomes increasingly complex, which obstructs or slows down innovation opportunities that result from effective communication. The mechanisms behind this relationship remain unclear, and should be examined to help understand the micro-foundations of secrecy. I propose that the restrictive dimensions of implementing secrecy as a knowledge protection instrument negatively impact internal knowledge sharing between employees. This leads to the following hypothesis:

(18)

18

In addition to the reduction of internal knowledge sharing of employees, purposeful knowledge sharing with individuals beyond the boundaries of the firm may be restricted. Innovating with external partners through intended knowledge spillovers provides value-creating opportunities that benefit the firm’s innovation capabilities (Laursen & Salter, 2006; Laursen & Salter, 2014). However, when knowledge flows are restricted, organizational learning and inter-firm cooperation becomes increasingly difficult for employees (Bos et al., 2015; James et al., 2013). Blocking external knowledge spillovers reduces unintended knowledge flows, however intended spillovers of knowledge may be likewise affected. Therefore, the second hypothesis is:

H2: Secrecy is negatively related to external employee knowledge sharing Figure 1 shows the conceptual model of this study.

H2

Figure 1. Conceptual model on the employee level H1

Secrecy Employee knowledge _sharing

Internal knowledge sharing

(19)

19

4. Methodology

4.1. Participants and procedures

To test the proposed hypotheses, I conducted an online survey at two Dutch multinational firms operating in the agricultural enhancement industry. To effectively measure the constructs, it was essential to engage with knowledgeable employees that operate with value-creating innovative activities and knowledge protection in their daily work as respondents. Knowledge is considered the most important resource to gain a competitive advantage in this industry, and the use of both patents and secrecy to protect innovations is prevalent. In collaboration with the senior managers at both firms project teams were selected based on their exposure to R&D and innovation processes.

A survey is an appropriate data collection method for this research, since it may provide insights into both individual perceptions and organizational policies (Baruch, 2008). I co-created the survey with a colleague at the University of Groningen, by use of web-based survey tool Qualtrics. The survey was published online in both the Dutch and the English language through an anonymous link. I decided on the use of a web-based survey tool compared to other distribution methods because of its efficient send-and-return systems, the automatic generation of datasets and overall convenience and verifiability (Kaye & Johnson, 1999). The data was collected from March until May 2019. In return for facilitating the research at both firms, I presented a report to the management teams to explain key insights and provide managerial suggestions. The survey consisted of 104 items and has been jointly developed with a colleague who examined knowledge sharing and the degree of dependence within teams in the field of Human Resource Management. In order to effectively measure the joint constructs, it was necessary to differentiate between project managers and their subsequent team members. Therefore, the survey originally existed in two versions: the first aimed at employees working in a supervisory role (e.g. project managers) which included 20 additional supervisor questions, and the second aimed at regular team members, from which those questions were omitted. The surveys underwent a meticulous translation process and several independent spelling and grammar checks to rectify errors that could have initiated misunderstandings and potentially unbalanced measurements. The survey was distributed through an anonymous link, which was highly valued by both firms. Additionally, the use of anonymity in the data collection reduces the level of socially desirable answers, which increases reliability of the research (Joinson, 1999).

(20)

20

For a full disclosure on the surveys as distributed at Firm A and Firm B, please refer to Appendix A. and Appendix B, respectively.

4.2. Data and Sample

We initiated the data collection process by contacting Firm A, which agreed to collaborate on the project through its established relationship with the University of Groningen. A first company visit was primarily informative, where a senior manager introduced the firm, questions were exchanged and a second meeting was scheduled. In the second meeting with the senior manager, we discussed the survey, purpose and timeline of the research in more detail. Beforehand, we provided a document stating the research details, purpose and preliminary survey questions to the firm. In this meeting, the details of conducting research at the firm were established. It was essential for the joint survey to identify which respondents belonged to the same team, which proved to be rather complex to set up. Employees were active in multiple ongoing project teams as either a project manager or regular team member, which, combined with the anonymity agreements, complicated the identification of teams in the research. To deal with this, we selected the optimal distribution of projects with a maximum amount of potential respondents that have minimal overlap, in cooperation with the senior manager. Accordingly, 13 project teams were identified resulting in a potential respondent base of 68 employees. The use of team numbers was subsequently explained in both the consent form and instruction email accompanying the survey, and participants needed to select their team number at the first question of the survey. Before receiving an overview of the team structures and being able to distribute the survey, we had to sign a non-disclosure agreement (NDA) to ensure that company information remained within the boundaries of this research project. While waiting on approval from the legal department, we finalized the survey based on feedback of both the senior manager and both thesis supervisors. When the NDAs had been received, signed and returned to the firm, the distribution of the survey had to be delayed due to unavailability of the senior manager. After several reminders, the survey was ultimately distributed through an instructive email that was forwarded to the selected employees by the senior manager. When the data collection reached the first deadline, an insufficient amount of respondents had completed the survey. Therefore, we decided to extend the deadline with a week, and we sent a reminder email that highlighted the relevance of the research and encouraged additional response. When after this period there still were insufficient respondents, we repeated the reminder process once more, which resulted in a total of 28 respondents of which many were impartial or unusable. The remaining sample size would not give an adequate reflection of the population, and thereby was insufficient to conduct valid statistical analysis.

(21)

21

resulted in collaboration negotiations with Firm B, and after a company visit and constructive meeting with two senior managers, it was agreed upon that I would be additionally conducting research at Firm B. By analysing the different departments and related division of employees, their potential respondent base was estimated at approximately 50 employees. Regarding the survey, at request of the management team the following adjustments were made: (a) the descriptive terminology of the existing survey was adjusted to meet firm terminology for accurate measurement (e.g. ‘project manager’ was adjusted to ‘department manager’), (b) an item determining the respondent’s department was added and (c) an additional firm-specific measurement scale that falls beyond the scope of this research was incorporated. The latter was included at the end of the survey to ensure the congruent validity of the research. To reduce complexity and stimulate a higher response rate, the two separate surveys for project managers and team members were merged into one survey that internally distinguishes between these two employee characteristics. This was established by including a question that showed related items when ‘yes’ was selected and skipped the supervisor-related items when ‘no’ was selected. By merging the two surveys in this way, I ensured that the reliability and validity of the survey were not affected. After implementing the senior management’s feedback and making final adjustments accordingly, the completed survey for Firm B consisted of 86 items and was distributed in May 2019, after drawing up and signing a confidentiality agreement. After two weeks of collecting data and one reminder several days before the final deadline, the data collection at the second firm resulted in 37respondents.

Ultimately, data was collected on a sample of 55 employees active in R&D intensive departments of the firms operating in the agricultural enhancement industry. Eight respondents utilized the English version of the survey, whereas the remaining 47 completed the survey in the Dutch language. For Firm A, out of a total of 68 potential respondents, the collection of 18 usable surveys resulted in a response rate of 26.5%. Considering the effort made, the periodical reminders and multiple postponements of the deadline to complete the survey, this response rate is rather low. For Firm B, out of a total of 58 potential respondents, 37 usable surveys were recorded, resulting in a response rate of 63.8%. The joint response rate of the firms subsequently amounts to 43.7% (55 out of 126 potential respondents).

4.3. Variables

(22)

22

they have either been constructed from existing firm-level measurements (Gold, Malhotra, & Segars, 2001) or incorporated by adapting them to the appropriate level. This alignment of data analysis with the level of theory allows for drawing relevant conclusions (Klein, Dansereau, & Hall, 1994). The questions regarding these variables were constituted by use of a Likert scale (Likert, 1932). The survey items regarding knowledge sharing, secrecy and the core control variables measuring patenting and workload were analyzed by factor analysis to estimate their aggregated effect.

4.3.1. Dependent variable. The first measurement of knowledge sharing consists of 14 items regarding the sharing of knowledge and skills between employees within and outside departments (Van den Hooff & De Leeuw van Weenen, 2004). These items were relevant since they include both knowledge sharing behaviour of individuals and the perceived knowledge sharing behaviour of their colleagues. Additionally, they differentiate between donating knowledge (gaining knowledge without asking for it) and collecting knowledge (gaining knowledge by asking for it) which allows for exploring if both types are affected equally by the independent variable. An example questions regarding individual knowledge sharing and knowledge collecting is “When I’ve learned something new, I tell my colleagues in my team about it” and an example of knowledge sharing behaviour of colleagues and knowledge collecting is “Colleagues outside of my team tell me what they know, when I ask them about it.” An additional measurement of knowledge sharing includes eight items that regard knowledge sharing of individuals in differing professional business settings (Silva & Odelius, 2018). Additionally, two items assessing the difficulty of sharing knowledge within and outside teams were included. The questions regarding knowledge sharing were answered on a 7-point Likert scale ranging from 1 = “strongly disagree” to 7 = “strongly agree”.

(23)

23

4.3.3. Control variables. I selected several control variables to be included in this research that may influence the effects of knowledge protection instruments on knowledge sharing, or the level of knowledge sharing independently. The first control variable is the use of patents as a knowledge protection mechanisms. Patenting is a relatively tangible and clearly defined mechanism for employees to implement. Even though disclosure of innovation-specific knowledge is limited to the requirements and a certain extent of secrecy upholds, it is easier for employees to share knowledge regarding patented innovations compared to sharing knowledge regarding innovations protected by secrecy. To explore the proposed effect of secrecy on knowledge sharing, patenting will act as a benchmarking variable to check its effects on knowledge sharing as a formal knowledge protection instrument. The measurement variable for patenting was established through the employee-level adapted construct of Gold et al. (2001) on knowledge protection, however applied to the concept of patenting instead of secrecy through three items. I developed three additional items that assess the involvement of employees in applying for patents, the perceived effectiveness of patents and its perceived importance for competitiveness. Example items regarding patenting are “The knowledge that I work with is protected by patents” or “I am convinced that patents are an effective way to protect knowledge”, and were answered on a 7-point Likert scale ranging from 1 = “strongly disagree” to 7 = “strongly agree”.

The second core control variable constitutes of workload. In this study, the subjective workload refers to the employee’s perception of having insufficient time to complete daily work tasks. A high workload may influence knowledge sharing negatively through a lack of time (Cleveland & Ellis, 2015; Siemsen, Roth, & Balasubramanian, 2008). To measure this construct, I included a measurement regarding workload by Van Veldhoven & Meijman (1994) (as cited in Janssen, 2001) in the survey. The measurement consists of eight items such as “Do you have too much work to do?” or “Do you work under time pressure?” which were answered on a 5-point Likert scale ranging from 1 = “never” to 5 = “always”.

(24)

24

establish networks and share knowledge. Therefore, an additional control variable which numerically measures the amount of teams in which respondents are active was included in the survey.

The fourth control variable refers to the occupational position of an employee within the firm. Michailova & Minbaeva (2012) indicate that levels of knowledge exchange may vary between different organizational positions, thereby discriminating between those employees currently active in a supervisory role and those that are not. Furthermore, respondents that have a supervisory role and are more involved in policy making are expected to be less influenced by the effects of their own policies through their inhabitation of higher perceived behavioural control (Hu, Dinev, Hart & Cooke, 2012). Therefore, I expect that employees that occupy a supervisor position will be less affected by the effects of secrecy on knowledge sharing. Correspondingly, the survey at Firm B included the dichotomous categorical item “Are you currently employed as department manager” with the options 1 = “yes” and 0 = “no” in distinguishing between respondent’s supervisory status. To incorporate the respondents at Firm A in this variable, this research quantifies the respondents that completed the supervisor survey and those that completed the team member survey, which were added manually to the variable.

Lastly, it is assumed that individual characteristics influence employee knowledge sharing behaviour (Xue, Bradley, & Liang, 2011). Riege (2005) states that from a knowledge management perspective there is general consensus on the influence of age differences, gender differences and differences in educational level on knowledge sharing. Previous research at large multinational firms additionally indicates that the emergence of subgroups in teams based on age or gender can negatively influence the levels of knowledge sharing within a firm (Gratton et al., 2007). Subsequently, Miller and Karakowsky (2005) find that the extent to which employees seek out information and feedback from their peers varies significantly between genders. Yang and Chen (2007) find educational level, to be significantly and positively correlated with the level of knowledge sharing. Therefore, I included age, gender and educational level as demographic control variables in the survey. Age was measured as a ratio variable, and gender as a dichotomous categorical variable with male (0) and female (1) as categories. The level of education was measured by including a categorical variable ranging from 1 = “primary school” to 7 = ´university PhD”.

4.4. Analysis

(25)

25

significant consequences for the study (Dong, 2013). To additionally validate the dataset I checked for outliers, since abnormalities in a small sample could substantially impact research outcomes. I used Tukey’s Hinges to detect outliers in the dataset. For team tenure one outlier at the 1.5 multiplier level was found. However, Hoaglin, Iglewicz, and Tukey (1986) argue that the appropriate multiplier for detecting outliers with this method is approximately 2.2. At the multiplier level of 2.2, the data point falls within the normal distribution range. Additionally, this relatively high value in the respondent’s team tenure was not considered an outlier through logic reasoning. The SPSS software indicated a variety of other outliers on the measurements of knowledge sharing, secrecy, patenting and workload. However, since variables were measured by use of Likert scales which encompass a predetermined set of answer values that provide representative data, this indication of outliers was disregarded.

(26)

26

valuable knowledge that must be appropriated through knowledge protection mechanisms. Regarding the organizational occupancy, 27% of the respondents holds a supervisor position, of which 67% is male and 33% is female.

Table 1 Descriptives

Firm A Firm B Total

N 18 37 55

Gender (male/female in %) 78/22 62/38 67/33

Age (M ± SD in years) 43.56 ± 12.67 41.83 ± 10.80 42.41 ± 11.36

Supervisor role (supervisor/no supervisor in %) 17/83 32/68 27/73

Level of education (%)

Primary school 0.0 0.0 0.0

High school 0.0 0.0 0.0

Secondary Vocational Education 5.6 8.0 7.3

Higher Vocational Education 22.2 27.0 25.5

University Bachelor 11.1 0.0 3.6

University Master 33.3 32.5 32.7

University PhD 27.8 32.5 30.9

University educational level (university/non-university in %) 72.2/27.8 65/35 67.2/32.8

PhD educational level (university PhD/non-university PhD in %) 27.8/72.2 32.5/67.5 30.9/69.1

The item measuring organizational tenure (Table 2) shows that respondents have been employed at their respective firms for 13.4 years on average (SD = 12.09). The shortest indicated organizational tenure is less than 1 year, and the longest 49 years. Additionally, the average involvement of employees within a team attributes to just over 8 years (SD = 8.42), whereas the shortest active member is engaged with a team for less than 1 year, and the longest active member for 33 years. On average, the respondents are active in just over 5 project teams simultaneously (SD = 3.00), with a minimum value of 1 and a maximum of 10.

Table 2

Tenure of respondents

Firm A Firm B Total

Organizational tenure (M ± SD in years) 13.15 ± 12.58 13.51 ± 12.02 13.40 ± 12.09

Team tenure (M ± SD in years) 2.38 ± 2.66 10.84 ± 8.88 8.07 ± 8.42

Involvement in different project teams (M ± SD in number of teams) 3.06 ± 1.26 6.00 ± 3.14 5.04 ± 3.00

(27)

27

level variable was coded into a university educational level dummy variable, which indicates if the respondent has at least a university degree (1) or another degree as their highest educational level (0). The variable was also recoded into the PhD educational level dummy variable to distinguish between respondents that have a doctorate degree (1) or not (0). Based on the career stages proposed by Wright & Bonett (2002) organizational tenure was recoded to the intermediate variable organization tenure level which consists of three levels: 1 = ”low” (0-2 years), 2 = “medium” (2-8 years) and 3 = “high” (> 9 years). Additionally, team tenure was recoded to the intermediate variable team tenure level based on the same criteria. Both newly coded variables were recoded as two sets of three dummy variables, indicating if the respondent belongs to the specific tenure level (1) or not (0). Hereby it will be more appropriate to make assumptions on the different stages of organizational tenure and team tenure and how they affect knowledge sharing.

The data of the dependent and independent variables is Likert scale data, which is considered ordinal and bound on a minimum and maximum value. It cannot be normally distributed, which could be problematic for parametric statistical analyses such as Pearson correlation tests, intraclass correlation and regression methods such as Ordinary Least Squares. However, Norman (2010) established that for samples that consist of Likert scales, these tests are robust and provide the same statistical significance as for normally distributed datasets, and may be used accordingly.

4.4.2. Factor analysis. To identify if the multi-item variables represent singular or multiple underlying dimensions, I conducted a factor analysis to estimate the aggregate effects of the items. According to Streiner (1994), a sample of at least 200 respondents and 5 entries per variable is required in order to conduct a significant factor analysis in survey research. Although this study does not adhere to these requirements, the sample size (N = 55) of above the absolute minimum of 50 is narrowly significant enough to examine the constructs and make careful assumptions about aggregation or interrelationships of variables. Before running the preliminary analyses, I reverse coded KS_23_IT “It is difficult to share knowledge within my team” and KS_24_E “It is difficult to share knowledge outside of my team”, since they were included as negative knowledge sharing statements instead of positive statements. Additionally, I reverse coded WL_5 “Can you do your work in comfort”, since a high value would measure a low workload, which is contradictory to the other workload variables. By reverse coding these variables, I certified that these variables are measured in the same direction as the others, and thereby can be included in statistical analysis.

(28)

28

in a total of 24 items on knowledge sharing. Based on their relevance for the model, I decided to separately run an EFA on the first 14 items by Van den Hooff & De Leeuw van Weenen (2004) and on the 10 items deducted from Silva & Odelius (2018). I examined the relatedness of variables to their underlying factors, which I extracted by use of Principle Axis Factoring (PAF). PAF is a reliable and appropriate method for this research, since it aims at obtaining an estimate of the factor structure within a total population and provides more natural results for theorizing on conceptual relationships (Norris & Lecavalier, 2010). I conducted these analyses using the direct Oblimin rotation, since the existence of correlations between the variables was expected. Oblique rotations such as Oblimin allow for correlation between factors and therefore provide a more realistic interpretation to a model or population than orthogonal rotations (Norris & Lecavalier, 2010).

To validate EFA as suitable method of analysis, I conducted Bartlett’s Test of Sphericity (BTS) for each subset of items to test if a sufficient correlation between variables exists. To verify the sufficiency of the sample size, I used the Kaiser-Meyer-Olkin test (KMO) and examined the average communalities of all EFAs, which should be at least .60 for small sample sizes (N < 100) (MacCallum, Widaman, Zhang, & Hong, 1999). For assessing the adequate number of factors, Fabrigar and Wegener (2011) advise that this choice is a statistical as well as a conceptual consideration. Therefore, I examined statistical criteria such as the Kaiser’s criterion and scree plots, and evaluated the significance of the factor outcomes for the conceptual model. In determining the relevant variables and the sufficiency of their factor loadings, I used the threshold of 0.40 as implied by Ford, MacCallum, & Tait, (1986). Subsequently, in checking for multicollinearity, I used a cut-off score of 0.80 to ensure that the variables are largely independent and allow for an accurate interpretation of the results (Field, 2013). Additionally, when items loaded on multiple factors with a value of .30 or higher, variables were excluded until this was no longer the case. To verify the models, I additionally examined the determinant of the correlation matrix which should exceed a value of 0.00001. For the extraction communalities I used a cut-off score of 0.20 (Child, 2006), since items with low common variance may indicate additional factors that fall beyond the scope of this study (Costello & Osborne, 2005). Additionally, I ensured that the explained variance of the accumulated factors in PAF exceeded 50% (Streiner, 1994). Lastly, to analyze the internal consistency between the indicator items, Cronbach’s alpha (α) was calculated for the aggregated factors.

(29)

29

examined their Variance Inflation Factors (VIFs). The dummy variables for high team tenure and high organizational tenure structurally caused multicollinearity problems with VIF factors exceeding the generally accepted maximum of 5 (Stine, 1995), and were exempted from analysis. None of the other combinations of variables in the regression models had problematic VIF values (maximal VIF value amounts to 3.63), which suggests that there are no further multicollinearity issues that would distort an appropriate estimate of the regression coefficients. The mean VIF values of the models have been indicated at the estimation results. Additionally, I checked for common method bias by use of Harman’s one-factor test. I found no indications of common method bias, since the principal component analysis for the variables indicated that no factor explains more than 15% of the total variance and there are seven model variables with an eigenvalue greater than one (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003).

(30)

30

5. Results

5.1 Factor analysis

I conducted EFAs on the subject-related survey items to aggregate correlating items into factors that measure their relevant underlying construct(s). Regarding the EFA on the first measurement (Van den Hooff & De Leeuw van Weenen, 2004), the KMO test was significant with a score of .64 at the .50 threshold (Hutcheson & Sofroniou, 1999), which is an acceptable score. The average communality of the items was .67, which is sufficient at the threshold of .60 (MacCallum et al., 1999). These tests indicated that the sample was large enough to conduct PAF. Additionally, the BTS showed a significant result (p < .001), which indicated that the correlation between the variables was also high enough to conduct PAF. A Pearson correlation test showed that the variables KS_13 and KS_14, have an extremely high correlation (r = .95, p < 001). Therefore, one should always be exempted from the analysis (Field, 2013), and I assume that the perception of transferring knowledge and skills substitutes for each other in this construct. Subsequently, KS_9 and KS_10, as well as KS_11 and KS_12 had problematically high correlations (r = .81, p < 001). These issues of multicollinearity were accounted for in conducting the factor analyses. After excluding several items that had high loadings across multiple factors, the factor structure with highest total explained variance and best conceptual fit was selected. The scree plot (Figure 2) indicates clear levelling off of its slope at four factors, which suggests to derive three factors. These factors all have eigenvalues above the Kaiser criterion (3.13; 1.72; 1.48). The factor loadings range from .61 to .90 and comply with the threshold of .40 (Ford et al., 1986). The communalities all exceed Child’s (2006) threshold of .20 and the determinant is sufficient (.02). The three factors consist of 8 indicators and cumulatively explain 79.3% of total variance which exceeds Streiner’s (1994) minimum of 50%. Their total (α = .76) and independent (α = .85; .78; .80) scores are reliable at the .70 threshold (Connely, 2011) and indicate internal consistency of the factors.

(31)

31

sufficient (.09). Their total (α = .81) and independent scores (α =.77; .81) indicate internal consistency and reliability of the factors (Connelly, 2011).

A total of 5 factors arise from the factor analyses that represent the underlying factors of knowledge sharing in further analysis. The factors were interpreted as (1) Team external knowledge sharing, (2) Team internal knowledge sharing, (3) Team internal skills sharing, (4) Organizational knowledge sharing and (5) Knowledge sharing network. These factors have been visualized in Table 3 and 4, including their factor loadings, eigenvalues and explained variance. The bold factor loadings indicate the highest loadings of the items.

Table 3

Factor loadings after rotation of the knowledge sharing items 1 (EFA)

1 2 3

1. Team external knowledge sharing

KS_4. When I’ve learned something new, I tell my colleagues outside of my team about it. .90 .03 -.15 KS_5. When they’ve learned something new, colleagues outside of my team tell me about it. .80 .04 .01 KS_6. Knowledge sharing with my colleagues outside of my team is considered a normal thing. .80 .25 -.00 KS_11. I share the information I have with colleagues outside of my team, when they ask me to. .61 -.26 .20

2. Team internal knowledge sharing

KS_3. Knowledge sharing with my colleagues within my team is considered a normal thing. .03 .85 .15

KS_2. When they’ve learned something new, colleagues within my team tell me about it. .03 .74 -.02

3. Team internal skills sharing

KS_8. I share my skills with colleagues within my team, when they ask me to. -.02 -.02 .85

KS_10. Colleagues within my team tell me what their skills are, when I ask them about it. .01 .12 .79

Eigenvalue 3.13 1.72 1.48

Explained variance (%) 39.17 21.54 18.54

Figure 3. Scree-plot after EFA on

knowledge sharing 2

Figure 2. Scree-plot after EFA on

(32)

32 Table 4

Factor loadings after rotation of the knowledge sharing items 2 (EFA)

1 2

4. Organizational knowledge sharing

KS_16. I use the knowledge derived from conversations with colleagues in the organization to solve problems at work

1.05 -.21

KS_15. During daily activities, I expose my knowledge by providing comparisons and examples .56 -.01

KS_17. I develop new ideas and concepts through dialogues with co-workers .54 .08

KS_19. I engage in conversations with colleagues on issues related to working during casual encounters (e.g. coffee breaks and organization gatherings)

.49 .25

KS_18. I share knowledge related to the activities I do in work meetings .42 .24

5. Knowledge sharing network

KS_21. I maintain interaction with groups or networks of people within the organization -.03 .81

KS_22. I maintain contact with specialists in the organization whom have recognized knowledge in specific subjects

.08 .80

Eigenvalue 3.32 1.11

Explained variance (%) 47.45 15.80

Secondly, to identify the descriptive categories of the independent variable secrecy, an EFA was conducted on the 12 items that measure the construct. Prior to interpreting the analysis, I identified if the conditions for conducting the PAF were met. The KMO test is significant with a score of .81, which is considered above average (Hutcheson & Sofroniou, 1999). The average communality of the items is .57, which is slightly below the recommended threshold of at least .60 but still reliable (MacCallum et al., 1999). The BTS was significant (p < .001), which allows for concluding that the sample size and the correlations between the variables are sufficient to conduct PAF. Initially, after excluding factors that loaded in multiple factors, three factors could be extracted. However, SEC_7 “I feel that the knowledge that I create is being valued and protected in my organization” loaded independently in a separate factor. Since a factor

consisting of a singular item is not relevant, the item was probably interpreted in a different way than intended, and therefore the factor was excluded from the EFA before re-assessment. The new scree plot (Figure 4) shows that two factors have eigenvalues above the Kaiser criterion (3.89; 1.50). The factor loadings range from .53 to .88 and comply with the threshold of .40 (Ford et al., 1986). There are no cases of multicollinearity at the cut-off score of .80 (Field, 2013), the communalities of all items

(33)

33

exceed Child’s (2006) threshold of .20, and the determinant is sufficient (.02). The two factors consist of eight indicators and cumulatively explain 67.4% of the total variance, which is sufficient according to Streiner’s (1994) minimum of 50%. The PAF on secrecy resulted in two factors that were interpreted as (1) Secrecy protection and, (2) Secrecy restrictions. The internal consistency of the total measurement (α = .82) and those of the two factors (α = .84; .75) supports the reliability of the factors Connelly, 2011). The two factors and their individual explained variance and eigenvalues are visualized in Table 5.

Table 5

Factor loadings after rotation of the secrecy items (EFA)

1 2

1. Secrecy protection

SEC_2. I use secrecy to protect knowledge from inappropriate use outside the organization .88 .02

SEC_3. I use secrecy because I am concerned that other organizations will copy the knowledge .76 .19

SEC_10. The knowledge that I work with is protected by secrecy .76 .14

SEC_11. It is clear to me which knowledge can be protected with secrecy .71 .11

SEC_1. I use secrecy to protect knowledge from inappropriate use inside the organization .53 -.15

2. Secrecy restrictions

SEC_6. I follow extensive policies for protecting sensitive information .03 .82

SEC_5. My use of email or other IT is limited because of concerns regarding information leakage -.07 .68

SEC_8. The organization has clearly identified the sensitive knowledge that is protected by secrecy .24 .56

Eigenvalue 3.89 1.50

Explained variance (%) 48.68 18.69

[EFA_PAT] Furthermore, in establishing the interrelatedness of the measurement of patents and include it as control variable, an EFA was conducted on the 6 items regarding patenting. The KMO test shows a significant value (.65) which

is considered an acceptable score (Hutcheson & Sofroniou, 1999). The average communality of the items (.61) lies above the threshold of .60 (MacCallum et al., 1999). Furthermore, the BTS was significant (p < .001), and thereby the necessary conditions for conducting a PAF were met. The scree plot (Figure 5) indicates that there are two factors that have an eigenvalue that adhere to the Kaiser criterion (2.96 and 1.32). Factor loadings range from .49 to .98 and are