Organizing for Open Innovation: Aligning internal structure with external knowledge search
Christoph Ihl a,*, Frank Piller a, Philipp Wagner a
RWTH Aachen University, Technology and Innovation Management Group
Kackertstrasse 7, 52072 Aachen, Germany Phone: +49 (0)241 - 80 - 96356 Fax: +49 (0)241 - 80 - 92367 Mail: ihl | piller | firstname.lastname@example.org * Corresponding author
Acknowledgements: The authors are thankful for valuable comments made by the organizers and participants of the special issue conference “Open Innovation: New Insights and Evidence” held at Imperial College London, participants of the “10th International Open and User Innovation Workshop” held at Harvard Business School, and participants of the EURAM 2012 conference held at Rotterdam School of Management. The first author is grateful for generous financial support by the “Stiftung Industrieforschung” and the “Peter Pribilla Foundation”.
Electronic copy available at: http://ssrn.com/abstract=2164766
Organizing for Open Innovation: Aligning internal structure with external knowledge search
In the past decade, research on open innovation has brought renewed attention to ways how firms can gain from the interaction with external sources of knowledge and innovation. Complementary internal management practices, however, that explain why some firms benefit from open innovation more than others are still largely unexplored. This study adopts the notion of open innovation as external knowledge search and investigates its mutual interdependence with internal organizational structures of a firm’s innovation function. Drawing upon behavioral theories about organizational search and information processing, we hypothesize how structural dimensions such as specialization, formalization and decentralization affect gains from open innovation. Based on a sample of German manufacturing firms, we find higher performance gains from open innovation by aligning internal organizational structures in terms of lower specialization as well as higher formalization and decentralization. These organizational contingencies of open innovation are further emphasized in light of firms’ internal R&D intensity: (1) Low specialization is especially beneficial for firms that try to align open innovation in a complementary fashion with their high internal R&D intensity. (2) Higher formalization and decentralization is essential for firms that try to substitute their low internal R&D intensity by the means of open innovation. Keywords: Open Innovation, innovative search, external knowledge, R&D intensity, organizational structure
Electronic copy available at: http://ssrn.com/abstract=2164766
Innovation has traditionally been located solely in the realm of firms’ internal activities, such as research & development. Yet, in light of recent advances of technology and significant changes in business conditions, Chesbrough (2006) has proposed that firms’ need to make more systematic or “purposive” use of external knowledge in order to increase innovation performance. In light of this observation, Chesbrough (2003) introduced the term open innovation to describe a concept in which firms increasingly access external sources of knowledge and technology (outside-in perspective) during their innovation process and bring in-house inventions to markets via external paths (inside-out perspective). The outside-in perspective of open innovation, which is the focus of our paper, refers to firms’ exchanges with a diversity of external knowledge sources (Laursen and Salter, 2006; Grimpe and Sofka, 2009). Extant research has shown that firms can benefit substantially from external knowledge integration and the utilization of a diverse set of external partners (e.g. Cassiman and Veugelers, 2006; Faems et al., 2010; Katila and Ahuja, 2002; Laursen and Salter, 2006). However, previous research also suggests that this kind of openness has limitations (e.g. Deeds and Hill, 1996; Duysters and Lokshin, 2011; Rothaermel and Deeds, 2006). These limitations likely arise due to additional costs and new governance challenges firms need to manage in order to appropriate external knowledge and learn from it. Recently, a number of scholars have called to investigate the internal capabilities, processes, and skills which determine the extent to which firms can learn from external knowledge sources (Dahlander and Gann, 2010; Van de Vrande et al. 2010). With our research, we respond to this call and focus on the organizational 1
prerequisites that characterize the challenge of learning. One aspect already frequently studied in this context is absorptive capacity, i.e. firms’ ability to recognize, assimilate and apply external knowledge for innovation (Cohen and Levinthal, 1990). Absorptive capacity has an obvious connection to open innovation and its performance effects (e.g. Laursen and Salter, 2006; West and Gallagher, 2006; Vanhaverbeke et al., 2007; Foss et al., 2011; Rothaermel and Alexandre, 2009; Spithoven et al., 2011). However, the concept of absorptive capacity bears considerable conceptual and empirical vagueness (Volberda et al., 2010). Therefore, it does not directly connect to open innovation and external knowledge search (Vanhaverbeke et al., 2007, p. 16). Most research approximates a firm’s absorptive capacity by its internal R&D intensity (Lane and Lubatkin, 1998). But internal R&D intensity can also be viewed as an alternative knowledge search strategy that is in conflict and need of coordination with external search strategies (Laursen, 2012; Laursen and Salter, 2006). Therefore, the objective of our study is to derive and investigate organizational dimensions of absorptive capacity (Jansen et al. 2005; Volberda et al., 2010; Foss et al. 2011) that are of relevance in managing open innovation. In particular, we aim to establish a more immediate and fine-grained theoretical connection between external knowledge sourcing on the one hand and organizational structures (Colombo and Delmastro, 2008; Burton and Obel, 2004; Miles and Snow, 1978; Mintzberg, 1979) on the other hand, based on behavioral theories about organizational search and information processing (Galbraith, 1973; Siggelkow and Levinthal, 2003). Taking a contingency perspective, we investigate (1) how firms should align their internal organizational structure with their respective levels of external knowledge search, and (2) how the effectiveness of this organizational alignment differs for high
and low levels of internal search and R&D, which likely pose different coordination requirements. Analyzing data from a large-sale empirical study of 365 manufacturing firms, our investigation adopts the concept of “fit” from a moderation or interaction perspective (Venkatraman, 1989). Our study emphasizes three variables constituting an organization’s structure: specialization, formalization, and decentralization (Mintzberg, 1979; Pertusa-Ortega et al., 2010; Volberda, 1998). These variables have already been shown to be meaningful in a broader context of knowledge search, learning, and innovation (Pertusa-Ortega, et. al. 2010; Rivkin and Siggelkow, 2003). However, the connection between these organizational practices and external knowledge sourcing in the understanding of open innovation has received only little attention in empirical research so far. Notable exceptions are mostly qualitative in nature (Bianchi et al., 2011; Chesbrough and Crowther, 2006; Dodgson et al., 2006; Sakkab, 2002). Foss et al. (2011) provide the only quantitative evidence we are aware of. However, their focus is on users as a single knowledge source and the mediating role of specific practices rather than on the moderating role of organizational structures. Our paper proceeds as follows. Section 2 presents open innovation in light of behavioral theories about firms’ organizational search and information processing. It derives hypotheses about how and when organizational structures of firms’ innovation activities are expected to impact the extent to which firms can benefit from open innovation and external search processes. Section 3 describes our sample, measures and methods used in the analysis. The empirical results are presented in Section 4. Section 5 refers to the discussion and implications of our findings. Section 6 concludes with a summary of contributions, limitations, and suggestions for further research.
Theory and hypotheses development
2.1. Open innovation from an innovative search perspective The exploitation and utilization of external knowledge and ideas in internal innovation processes can be regarded as the core of the open innovation paradigm (Laursen and Salter, 2006). Activities with regard to connecting and exploiting external knowledge sources can be defined as a firm's knowledge search strategies (Sofka and Grimpe, 2010). In other words, open innovation activities find their expression in the specification of search strategies. Knowledge search is conceptualized as an activity by which organizations solve problems and attempt to recombine knowledge for the objective of generating new products (Katila and Ahuja, 2002). By engaging in knowledge search, firms expand and renew their knowledge (base), which puts them in a position to be more innovative and successful (Levinthal and March, 1981; Rosenkopf and Nerkar; 2001). Knowledge search is one of the central aspects for the comprehension of innovation success (Nelson and Winter, 1982; West, 2000). Knowledge search processes require a lot of resources in terms of time, skills, and financial resources (Cohen and Levinthal, 1990; Levinthal and March, 1993). Therefore, search activities may be constrained by the alternatives considered, as organizations and management may suffer from cognitive limitations (Ocasio, 1997; Gavetti and Levinthal, 2000). Accordingly, search processes of an organization are often very localized, which means that the organization's members search along trajectories, within known fields, and with regard to knowledge they already are familiar with (Stuart and Podolny, 1996). In order to stay competitive and constantly renew their knowledge base, however, organizations need to overcome these local
search tendencies (Leonard-Barton, 1992; Rosenkopf and Nerkar, 2001; Stuart and Podolny, 1996). Research has shown that it is beneficial for firms to differentiate their search efforts and to engage in more distant as opposed to local search (Laursen, 2012). This differentiation may relate to the distance of targeted technological fields (Katila and Ahuja, 2002), but also to organizational boundaries (Rosenkopf and Nerkar, 2001). External knowledge search beyond organizational boundaries, as opposed to internal search within a R&D department, has been shown to enhance innovation performance strongly (Laursen and Salter, 2006; Leiponen and Helfat, 2010; Rothaermel and Alexandre, 2009). Hence, research on external and internal knowledge search provides a theoretical and empirical foundation of open innovation. However, external knowledge search does not come free of limitations either. Earlier research has found that external knowledge search and sourcing has an inverted Ushaped effect on innovation performance (Faems et al., 2010; Katila and Ahuja, 2002; Laursen and Salter, 2006; Rothaermel and Deeds, 2006; Rothaermel and Alexandre, 2009; Deeds and Hill, 1996; Duysters and Lokshin, 2011). Search activities and the relations to the respective external sources need to be managed, and the acquired knowledge needs to be processed by the organization in order to exert innovation impact. The constraints firms face with regard to their processing capacities mainly derive from the restraints of attentive resources and the limitations of operational absorption capacities. Constraints may also result from a necessary coordination and balance with internal search in the form of R&D. Although there is evidence that external and internal knowledge search are complementary in some contexts (Katila and Ahuja, 2002; Cassiman and Veugelers, 2006; West, 2000), Laursen and Salter (2006) find that firms with high internal R&D intensity have
difficulties in profiting from external knowledge search. This apparent trade-off also is guiding our research in this paper. 2.2. Organizational contingency of open innovation Organizational structure has been shown to impact firms’ effectiveness regarding the communication and processing of information (Galbraith and Nathanson, 1978; Mintzberg et al., 2003; Olson et al., 1995). It has also been connected to the ability of a firm to innovate (Argyres and Silverman, 2004; Damanpour, 1991; Tidd et al., 1997), to absorb, proceed upon, and learn from external knowledge (Jansen et al., 2005; Van den Bosch et al., 1999), and relate to external parties (Lane and Lubatkin, 1998). These aspects all represent ingredients for successful open innovation, yet, the question remains how the organizational structure implemented by a firm supports its open innovation activities directly. We argue that firms need to align their organizational structure according to the desired level of external knowledge search and internal R&D intensity in order to manage the constraints of opening their innovation processes, as outlined in the last section. Research has suggested that external knowledge can only be utilized successfully when firms manage to modify their organizational structure to facilitate open innovation (Bianchi et al., 2011; Dahlander and Gann, 2010). The potential to process information between internal units and the external environment is to a large extent determined by firms' organizational structure (Cohen and Levinthal, 1990; Van den Bosch et al., 1999). This highlights the importance of a firm's structural composition for successful knowledge integration and innovation. The notion of alignment can be described by the concept of “fit” which is largely based in the domain of contingency theory (Venkatraman, 1989). Contingency theory proposes that there is not one right way to design an organization. Rather, the fit of certain 6
organizational structures depends on given contexts or contingency factors (Galbraith, 1973). Accordingly, we consider a firm's mix of external knowledge search and internal R&D intensity as the contingency factor for the appropriateness of certain organizational structures by which innovation performance then is jointly determined. As such, we follow the interaction approach, or the bivariate interpretation of fit, that assumes fit to be “an interaction effect of organizational context and structure on performance” (Drazin and Van de Ven, 1985: 515). We extend fit to three variables in a regression-based analysis with moderator variables, interactions, and sample splits (Venkatraman, 1989). While this conceptualization of alignment and fit has limitations compared to systems, multi-contingency or configurational approaches (Birkinshaw et al., 2002; Burton et al., 2002; Fiss, 2007), it can be seen as a first step towards more comprehensive understanding of organizational configurations (Venkatraman and Prescott, 1990). Organizational learning is largely dependent on a firm's contacts with external knowledge sources (Lane and Lubatkin, 1998). As a consequence, organizational search and a firm's openness towards external sources are seen as important mechanisms for organizational learning (Lane et al., 2006). In this concern, several studies have investigated the influence of organizational structure on a firm's search behavior (Cassiman and Valentini, 2009; Siggelkow and Levinthal, 2003; Zhang et al., 2007). De Boer et al. (1999) suggest that organizational structure embodied in basic organizational forms affects a firm's ability to integrate external knowledge. Besides the ability to identify and source external knowledge, organizational learning is shaped by a firm's ability to link external and internal knowledge (Bessant and Venables, 2008). Cohen and Levinthal (1990) refer to this capability as the inwardlooking component of a firm's absorptive capacity, and highlight its importance for an 7
effective organizational learning, as it facilitates efficient internal knowledge processing mechanisms (i.e. knowledge sharing). In this regard, previous studies have put forward the importance of organizational structure for inter-unit knowledge sharing (Tsai, 2002; Willem and Buelens, 2006), because it affects internal communication processes (Guetzkow, 1965), and also knowledge management (Lam, 2000; Pertusa-Ortega et al., 2010). Following the argument of Cohen and Levinthal (1990), external search strategies remain ineffective without the ability of the firm to communicate and share internally what has been absorbed from the environment. In other words, even if a firm successfully manages to search for knowledge externally and to establish and maintain linkages to external knowledge sources, the firm will not be able to achieve high levels of innovation performance in the absence of internal knowledgeprocessing capabilities. Organizational structures facilitate learning and innovation by balancing different static and dynamic elements. However, they overlap in the most fundamental and prevalent dimensions of specialization, formalization, and (de-) centralization (Burton et al., 2002; Burton and Obel, 2004; Miller and Dröge, 1986; Mintzberg, 1979; Olson et al., 2005; Volberda, 1996, 1998; Vorhies and Morgan, 2003; Walker and Ruekert, 1987). According to Walker and Ruekert (1987), using these dimensions one can resemble ideal-type organizational designs similar to those proposed by Burns and Stalker (1961) and their distinction of mechanistic and organic organizations, or the “organizational archetypes” identified by Mintzberg (1979). Hence, our study uses the three central structural dimensions of specialization, formalization, and decentralization to investigate the organizational contingencies of open innovation.
2.3. The moderating effect of specialization “Specialization” or “differentiation” refers to the division of tasks and activities into subtasks and the assignment of these tasks to and only to specific members or units of the organization as their prime activity (Mintzberg, 1979; Mintzberg et al., 2003). In firms exhibiting high degrees of specialization increased division of labor creates groups of specialists who direct their efforts to a well-defined but limited range of activities (Ruekert et al., 1985). In such an environment, tasks are "performed by someone with that function and no other" (Pugh et al., 1968: 73) Specialization enhances a firm's knowledge performance due to accumulation and mastery of certain skills and abilities of specialists within their specific range of functions (Willem and Buelens, 2006; Pertusa-Ortega et al., 2010). However, Damanpour (1991) and Damanpour and Gopalakrishnan (1998) suggest that the positive effect of specialization refers to later stages of an innovation process. Hence, specialization may be associated with certain drawbacks in the compatibility with efforts to organize a distributed external knowledge search. An increased generation of domain-specific knowledge due to high levels of specialization involves the development of different languages and views between the various subunits of a firm (Grant, 1996). Hence, specialization leads to decreasing differences within and increasing differences between subunits (Willem and Buelens, 2006). Although specific knowledge within multiple subunits indicates desirable knowledge heterogeneity at the firm-level, it is also associated with increasing structural and mental boundaries inside the firm (Olson et al., 2005; Pertusa-Ortega et al., 2010). An increase of inter-unit boundaries may increase the costs of communication and learning (Colombo and Delmastro, 2008), and thus inhibit knowledge transfer (Willem and Buelens, 2006). Since boundaries between subunits constitute interfaces across
which knowledge is transferred within a company, and as each interface bears the risk of potential knowledge loss, a firm may not be able to fully leverage the potential of its external search strategy when specialization is pushed too far. Garud and Karaswamy (1995: 98) speak of the danger of units “hoarding knowledge”. This could constitute a major hurdle for a firm to benefit from external knowledge integration, which requires a firm-wide dissemination of knowledge and thus learning processes between subunits (Stieglitz and Billinger, 2007). Van den Bosch et al. (2003) suggest that specialization, thus the division of tasks, equally leads to the division of knowledge. In turn, enormous efforts may be required for re-integrating diverse knowledge-based activities. Finally, Cohen and Levinthal (1990) argue that specialization undermines innovation performance by reducing diversity, which may be a prerequisite for accessing and absorbing new knowledge. Low diversity might be indicated by external search being restricted only to specialists, thereby reducing the variety of the search process. Summarizing, though specialization may facilitate external search and may be conducive to the application of acquired knowledge, it may likewise limit the variety in external knowledge sources, as specialists tend to pursue “narrow” search endeavor (Van den Bosch et al, 2003). Specialization also poses significant strains on the transfer and application of knowledge. Hence, the following hypothesis is stated: Hypothesis 1: Specialization negatively moderates the effect of external knowledge search on innovation performance.
2.4. The moderating effect of formalization Formalization is defined as the degree to which roles, authority relations, instructions, norms and sanctions, ways of communication, and procedures are defined by rules (Child, 1972; Khandwalla, 1977). Hence, the level of formalization reflects the degree of freedom members of an organization have in pursuing tasks and in establishing intra- and inter-firm relationships (Argouslidis and Baltas, 2007; Ruekert et al., 1985). Formalization is often measured by the existence of job descriptions, manuals, or control arrangements (Damanpour, 1991; Miller and Dröge, 1986). The clear definition of rules and the assignment of distinct methods and procedures to functional roles in an organization yield the development of adeptness in a limited area of activities. This in turn results in lower error rates and higher process efficiency (Hage, 1965; Ruekert et al., 1985). Furthermore, formalization codifies best practices and provides organizational memory that facilitates the diffusion of organizational capabilities (for instance, capabilities of how and when to tap into external knowledge sources) as well as the application and the transfer of knowledge (Levitt and March, 1988; Lin and Germain, 2003; Pertusa-Ortega et al., 2010). On the other hand, the strong emphasis on rules and procedures may lead to unchanging patterns of action and reduces process flexibility, as it hinders individuals from deviating from established behavior (Weick, 1979). This might constitute a substantial drawback, since flexibility has been found to facilitate innovation processes (Aiken and Hage, 1971; Damanpour, 1991). Scholars have argued that firms might risk ignoring important innovation stimuli (Jansen et al., 2006), because formalization directs attention only towards restricted aspects of the firm's external environment, subsequently reducing the firm's scope of knowledge search (Jansen et al., 2005; Weick, 1979). 11
Yet, it has also been argued that the existence of norms and explicit procedures could facilitate a firm's ability to identify and integrate external knowledge (Jansen et al., 2005; Vega-Jurado et al., 2008). Without formalization, external search and integration would suffer from being “disorganized, sporadic or ineffective” (Okhuysen and Eisenhardt, 2002: 383). Pertusa-Ortega et al. (2010) see formalization as a means to reduce ambiguity by “behavioral directives” rather than clear specifications. The reduced ambiguity enhances a firm’s ability to utilize external knowledge. Formalization offers the necessary procedures that facilitate communication with specific knowledge sources, and endows firms with the competence to access knowledge from these sources (Vega-Jurado et al., 2008). Formalization may also improve a firm’s capacity to apply the knowledge. Through formalization, “guidelines” for communication and exchange can be established, thereby improving cooperation among employees and units (Cordón-Pozo et al., 2006). Cordón-Pozo et al. (2006) further argue that formalization also helps to transfer knowledge between units, as it lays out norms and procedures for engaging in such an exchange. Additionally, formalization was found to motivate employees to share explicit and tacit knowledge (Dyer and Nobeoka, 2000; Jansen et al., 2005). Formalization is said to be similar to routinization (Feldman and Pentland, 2003), which is argued to enable flexibility (Becker et al., 2005) and reduce ambiguity (Pertusa-Ortega et al., 2010). This, in turn, may be beneficial for dealing with contingencies, engaging in experimentation, and creating new knowledge (Adler and Borys, 1996). Thus, the following hypothesis is proposed: Hypothesis 2: Formalization positively moderates the effect of external knowledge search on innovation performance.
2.5. The moderating effect of decentralization The degree of decentralization reflects the locus of decision-making power and refers to whether decision authority is concentrated or rather dispersed in an organization (Pfeffer, 1981). When decision-making authority is closely held by top managers (i.e. concentrated) the organizational structure is “centralized”. In contrast, “decentralized” structures exhibit a high degree of participation in decision-making, since decision rights are delegated to middle- and lower-management levels (Aiken and Hage, 1971). Accordingly, centralization and decentralization are opposite ends of the same scale (Olson et al., 2005). Scholars have asserted that centralization may have a positive effect on innovation especially under dynamic environmental conditions (Adler and Borys, 1996; Gupta et al., 1986). Top-down directives, for example, offer clear lines of communication and involve unambiguous responsibilities. However, Colombo and Delmastro (2008) identified various sources of organizational failure due to concentrated decision authority, such as the occurrence of information transmission leaks and delays, the distortions of intra-firm communication, as well as information overload due to narrow communication channels. Due to increased levels of employee participation along all processes, decentralized organizations, on the other hand, tend to generate a higher variety of innovative ideas (Damanpour, 1991; Ullrich and Wieland, 1980). The dispersion of decision rights to middle- and lower-management levels enables the formation of sub coalitions and has a positive effect on the number of possible promoters of innovative projects (Thompson, 1965). Accordingly, the literature proposes a rather negative relationship between centralization and innovativeness (Aiken and Hage, 1971). Scholars have suggested that decentralization facilitates knowledge integration and knowledge sharing. Foss 13
et al. (2011), for example, investigate how established firms can adapt organizational structure and practices in order to more efficiently leverage user and customer knowledge. The authors argue that decision rights should be collocated with those employees who are best informed about what decision is appropriate in a given context. Also Ruekert, Walker and Roering (1985) argue that decentralization is likely to be beneficial for innovating companies, since it empowers those employees who are close to the issue to make decisions and to implement them rapidly. Cohen and Levinthal (1990) suggest that decentralization increases the number of potential recipients of external knowledge, thus increasing the number of interfaces of a firm with its external environment. Decentralization enhances the adoption of new attitudes and behaviors (Pertusa-Ortega et al. (2010), which is important for external knowledge use. With respect to knowledge sharing, previous research has suggested that decentralization increases the willingness to share knowledge (Gupta and Govindarajan, 2000). Additionally, decentralization broadens internal communication and improves the quality of knowledge sharing among subunits (Sheremata, 2000; Van Wijk et al., 2008). Finally, broad external search comes along with great requirements for management to allocate their attention to individual tasks and members (Laursen et al., 2007). Yet, this managerial attention is a scarce good (Ocasio, 1997). Decentralization provides the opportunity for delegation to lower management levels. Again, decentralization may be conducive as investment needs may be best known on levels close to the external source. All in all, we propose: Hypothesis 3: Decentralization positively moderates the effect of external knowledge search on innovation performance.
We further seek to qualify these proposed moderating effects by exploring their direction and strength for high and low levels of internal R&D intensity. The motivation for this lies in the contradictory findings of prior research on open innovation. There is evidence that external and internal knowledge search are complementary in some contexts (Katila and Ahuja, 2002; Cassiman and Veugelers, 2006; West, 2000). Also Laursen and Salter (2006) follow an absorptive capacity argumentation to propose a complementary relationship between external knowledge search and internal R&D. However, they find a substitutional relationship in the sense that firms with high internal R&D intensity have difficulties in profiting from external knowledge search, and only firms with low internal R&D intensity can achieve some compensation by the means of external search. In a post-hoc explanation, Laursen and Salter (2006) conclude that the intensity of internal R&D may spur a “notinvented-here”-attitude and a reluctance to adhere to external stimuli for innovation. Based on this finding, internal R&D has likely to be viewed as an alternative search strategy that may be in conflict and need of coordination with external search strategies (Laursen, 2012). In the empirical part of this paper, we therefore also explore how the effectiveness of the proposed organizational alignment with external search differs for high and low levels of internal R&D, which likely pose different coordination requirements.
Data and methods
3.1. Sample description The empirical analysis is based on cross-sectional survey data which was collected from February to April 2010. The target population of 3,709 German firms was
selected from the Amadeus database, a product offered by Bureau van Dijk, based on being in the manufacturing sector and on availability of sufficient secondary data on firm size and financials. The selected firms were contacted via telephone in order to identify a contact person responsible for innovation-related activities and with a good overview of the respective firm’s organization and strategy (e.g. head of R&D departments or leading innovation managers). Managers who were willing to participate received personalized emails containing a link leading to the online questionnaire. The questionnaire was pretested and based on the logic of the Eurostat Community Innovation Survey (CIS) to ensure interpretability of results in light of existing research (Cassiman and Veugelers, 2006; Laursen and Salter, 2006; Leiponen and Helfat, 2010). To increase the response rate, a reminder e-mail was sent out three weeks after the initial e-mail was sent. This was accompanied by follow-up phone calls to the firms which had initially indicated willingness to participate in the survey but had not responded up to that point. A final reminder email was sent out three weeks after that. We received a total of 676 responses of which 382 responses were fully completed. 12 firms identified themselves as non-innovators. This proportion was considered as being too small to econometrically correct for a selection bias, so we excluded these observations from the analysis. Another 5 observations had to be excluded due to missing data on secondary variables obtained from the Amadeus database. Thus, 365 firm observations are included in the analysis. The resulting response rate of around 10% has to be considered acceptable, given respondents from higher management levels to an online survey (Klassen and Jacobs, 2001). To assess representativeness and non-response bias for our sample, we compared (1) the sampled firms with the non-respondents based on observable indicators from
the Amadeus database, and (2) early and late respondents within the sample. While step (2) did not yield significant differences, step (1) revealed that our sampled firms are larger compared to the non-respondents. All survey questions regarding external knowledge search and organizational practices refer to the average of a three-year period from 2007 to 2009. The innovation outcomes were asked to be evaluated only for the last year of that period. We attempted to temporally separate the independent and the dependent variables in order to account for time lags between innovation activities and innovation outcomes. This also adds to overcoming potential common method bias concerns (Rothaermel and Alexandre, 2009). 3.2. Measures The dependent variable of innovation performance was measured by the fraction of a firm’s 2009 turnover stemming from products that had been (1) newly introduced to the market during the period 2007-2009, (2) new to the firm but not new to the market, and (3) significantly improved during the period 2007-2009. These three fractions were added to derive an overall measure of innovative sales. The key independent variable of openness was measured by the number of external knowledge sources firms used in their innovation activities during the period of 20072009. The list of external sources include: (1) other internal units, (2) suppliers, (3) customers, (4) competitors, (5) private research institutes and commercial laboratories, (6) universities and other higher education institutions, (7) public research institutes, (8) consultants and open innovation intermediaries, (9) public information (e.g. patens, industry-specific literature, scientific publications, company reports etc.), (10) official events (e.g. exhibitions and fairs, professional workshops
and conferences, trade associations etc.). The internal reliability of these 10 indicators was sufficient (α=0.74). The first structural measure of specialization was adapted from Gibson and Birkinshaw (2004) and Volberda (1996, 1998). It contained two items which were measured in 5-point Likert scales ranging from strongly disagree to strongly agree: (1) “Innovation activities in our company were separated into different functional areas (e.g. according to basic versus applied research; different products or regions)”; (2) “Innovation activities in our company were structurally separated from other functions (e.g. Marketing, Sales, Production)”. The internal reliability of these 2 items was sufficient (α=0.81), so that they were averaged and the results was standardized. The structural measure of formalization was adapted from Desphandé and Zaltman, (1982) and Sommer et al. (2009). It contained two items which were measured in 5point Likert scales ranging from strongly disagree to strongly agree: (1) “Innovation activities in our company were based on strict process steps and detailed task descriptions.”; (2) “Innovation activities in our company are based on detailed process planning and risk analysis”. A third item was constructed based on the usage of 16 different formalized tools that are relevant in the planning and controlling of innovation projects (e.g. patent analyses, NPV analysis, benchmarking, etc.). The internal reliability of these 3 items was sufficient (α=0.68), so that they were normalized and averaged. Then the result was standardized. Finally, the measure for decentralization was adapted from Hitt and Brynjolfosson, (1997) and Kretschmer and Mahr (2010). On the one hand, it measures the extent to which decision-making power is delegated to lower hierarchical levels in the organization regarding the following issues: (1) The prioritization of innovation 18
projects, (2) the coordination of innovation projects, (3) the allocation of specific innovation tasks, (4) the utilization of specific innovation methods, procedures and instruments. Respondents had to rate on a 4-point scale representing the follwoing hierarchical levels: 1 = team members, 2 = team leader, 3 = head of department, 4 = top management level. The internal reliability of these 4 items was sufficient (α=0.89), so that they were averaged and then standardized. On the other hand, decentralized structures may only be fruitful if employees not only have the decision rights, but are also motivated and empowered to execute them. Hence, an additional item was constructed based on the usage of 10 HRM practices to stimulate innovation within firms by setting incentives and providing sufficient training and experience (cf. Laursen and Foss, 2003). Following formative measurement logics, we averaged the two components of decentralization and then standardized the result. We further added the following control variables: (1) R&D intensity was measured by the natural logarithm of R&D expenditures over sales per year on average during the period of 2007-2009. (2) Environmental uncertainty in terms of technological change and customer demands as well as equivocality in terms of technical complexity and uncertainty of innovation projects was measured by two items each, because these variables have been argued to influence organizational requirements with respect to information processing (cf. Daft and Lengel, 1986). (3) Slack resources were measured by the logarithm of cash resources averaged for the years 2006-2008. (4) Performance feedback was measured by the difference of a firms’ return on assets minus their performance aspirations averaged over the three years from 2006-2008. Aspiration in turn was measured as the weighted average of own historical performance and the average performance of firms with the same two digit NACE code within our target population of 3709 firms. These financial measures have been
shown to influence firms’ search behavior (cf. Greve, 2007) and were obtained from the Amadeus database. (5) Firm size as the number of employees and firm age in years were obtained from the Amadeus database and included as natural logarithm. (6) Finally, industry dummies were created on the basis of grouping NACE codes from the Amadeus database. Table 1 shows the descriptive statistics for the variables used in this study.
------------------ Insert Table 1 about here --------------------
3.3. Analysis The dependent variable in our model is double censored (having an upper censoring point at the score 100 and a lower censoring point at scores that equal 0). The most appropriate method to account for this censoring is a Tobit analysis (Greene, 2003). Our measure for innovation performance is skewed to the left, which compromises the underlying assumption of normally distributed residuals in the Tobit model. To account for this departure from the normality assumption, we employed a logarithmic transformation of the dependent variable. We employ interaction terms and / or median splits in the regression analysis both of which has been used before in similar studies (Laursen and Salter, 2006; Rothaermel and Deeds, 2006). We do not solely rely on judging and comparing size and significance of coefficients, which has been argued to be difficult in non-linear models in general and in presence of interaction terms or subsamples with different scaling (i.e. error variances) in particular (cf. Ai and Norton, 2003; Long, 2009). Hence, we conduct a series of postestimation tests and procedures: (1) we calculate Wald tests on the difference of
coefficients from different equations; (2) we calculate Wald type tests on the difference between inflection points of inverted U-shape relationships. We interpret a significant shift to the right as an increase in a firm’s capability to manage high number of external knowledge sources (Rothaermel and Deeds, 2006). This test compares the zero of the partial derivatives of inverted U-shape relationships in two subsamples and involves the ratio of two coefficients, which should also resolve the scaling issue between different subsamples. (3) We compute predictions and marginal effects, which are not affected by scaling, for different levels of openness and the moderator variables. Instead of setting the other variables to their means, we compute these effects for each observation and then average the results. (4) Finally, these average predictions and marginal effects are compared in terms of significant differences following the procedure suggested by Long (2009), in order to assess significant upward-shifts or in increase in slopes. The standard errors for the Wald test, predictions and marginal effects are obtained using the delta method (Greene, 2003).
Table 2 shows the first set of Tobit regression results. The base Model 1 confirms Laursen and Salter’s (2006) hypothesis of an inverted U-shape relationship between openness and innovation performance, which is further qualified in the following analysis. Model 2 and 3 compare this relationship between firms with low and high degrees of specialization. A comparison of regression coefficients reveals that the effect of openness is only significant for firms with low specialization. While a likelihood-ratio comparison of all coefficients simultaneously is not significant
(χ2(15)=16.97, p=0.32), a Wald test for single coefficients does reveal a significant difference for the linear term (diff=0.91, s.e.=0.44, p=0.02) and the squared term (diff=-0.07, s.e.=0.04, p=0.09) of openness. The Wald-type test for calculating the inflection points shows a significant inflection point at 7.20 knowledge sources (s.e.=0.45, p=0.00) for firms with low specialization. But it yields insignificant results for firms with high specialization (13.69 sources, s.e.=25.32, p=0.59), indicating that the relationship does not have a significant curvature.
------------------ Insert Table 2 about here --------------------
To further qualify these differences, we take a look at average predictions and marginal effects. Panel A in Figure 1 shows that firms with low specialization yield a higher innovation outcome than firms with high specialization for high level of openness, i.e. between 4 and 9 knowledge sources. Panel B shows that this difference is significant at least in the range of 5.5 to 8 knowledge sources. Panel C shows the average marginal effects of openness in innovation outcome, which is insignificant for firms with high levels of specialization over the entire range of openness. Hence, firms with high specialization do not seem to be able to benefit from open knowledge search. For firms with low specialization we find further evidence for an inverted U-shape indicated by marginal effects ranging from significant positive to significant negative.1 Accordingly, Panel D shows the significant differences in marginal effects between low and high specialization firms. Finally, Model 10 in Table 3, where specialization is introduced via interaction terms, also
For the sake of visual clarity, we do not include confidence intervals in graphs that show predictions and marginal effects of more than one group or subsample, but rather describe the ranges of significance verbally. These results are available upon request.
provides evidence for a significant moderation such that specialization attenuates the inverted U-shape relationship of openness on innovation outcome. All in all, we conceive this as support for Hypothesis 1 that specialization negatively moderates the effects of external knowledge search. On the basis of the empirical evidence up to now, we can say that high levels of internal specialization are detrimental for firms that want to benefit from high levels of external search.
------------------ Insert Figure 1 about here --------------------
Next, we compare firms with low and high levels of formalization in Models 4 and 5. This time, the coefficients referring to openness are significant for firms with both low and high degrees of formalization, albeit higher in magnitude for high formalization. A Wald test reveals that these differences are significant neither for the linear term (diff=-0.52, s.e.=0.59, p=0.38) nor for the squared term (diff=0.03, s.e.=0.04, p=0.46) of openness, while the entire coefficient vector is significantly different (χ2(15)=47.75, p=0.00). Wald type tests show significant inflection points of 6.91 knowledge sources (s.e.=0.77, p=0.00) for firms with low formalization and of 7.60 knowledge sources or firms with high formalization. This indication of a higher open innovation capability of firms with higher formalization, however, is not significant (diff=0.86, s.e.=0.86, p=0.43).
------------------ Insert Figure 2 about here --------------------
Figure 2 displays average predictions and marginal effects for firms with low and high formalization. Panel A shows that firms with high formalization yield a higher innovation outcome than firms with low formalization already beginning with moderate levels of openness, i.e. from 3 up to the maximum of 10 knowledge sources. Panel B shows that this difference is significant from 5 knowledge sources onwards. Panel C shows the average marginal effects of openness in innovation outcome, which is positively significant for firms with low levels of formalization up to 5 knowledge sources and then turns insignificant. For firms with high levels of formalization we find evidence for an inverted U-shape indicated by marginal effects ranging from significantly positive to significantly negative. Panel D shows that the positive marginal effects of openness are significantly higher for firms with high formalization compared to firms with low formalization in the range of approximately 1 to 6 knowledge sources. Model 13 in Table 4, where formalization is introduced via interaction terms, does not provide evidence for a significant moderation when judged on the basis of significant interaction terms. However, when Model 13 is used to calculate average predictions and marginal effects, the results show a significant upward shift due to formalization that is substantially identical to that obtained from Models 4 and 5. This result likely amounts to the ambiguity of testing moderating hypotheses via interaction terms in non-linear models as discussed above. All in all, we conceive this as support for Hypothesis 2 that formalization positively moderates the effects of external knowledge search. On the basis of the empirical evidence of an upward shift in the relationship between openness and innovation outcome, we can say that high levels of internal formalization are beneficial for firms such that they are able to gain more from a given (high) number of knowledge sources. There is no evidence that firms
with high formalization are able to manage more knowledge sources, what might have been indicated by a higher inflection point. Models 6 and 7 provide the basis for the comparison of firms with low and high levels of decentralization. The coefficients referring to openness are significant for firms with both low and high degrees of decentralization, albeit higher in magnitude for high decentralization. A Wald test reveals that these differences are significant neither for the linear term (diff=-0.31, s.e.=0.55, p=0.57) nor for the squared term (diff=0.01, s.e.=0.04, p=0.84) of openness, while the entire coefficient vector is significantly different (χ2(15)=27.31, p=0.03). Wald type tests show significant inflection points of 6.52 knowledge sources (s.e.=0.75, p=0.00) for firms with low decentralization and of 8.30 knowledge sources or firms with high decentralization. This difference in inflection points is marginally significant (diff=1.79, s.e.=0.98, p=0.07) and hence indicates a higher open innovation capability of firms with higher decentralization.
------------------ Insert Figure 3 about here --------------------
Figure 3 displays average predictions and marginal effects for firms with low and high decentralization. Panel A shows that firms with high decentralization yield a higher innovation outcome than firms with low decentralization beginning with approximately 4.5 up to the maximum of 10 knowledge sources. Panel B shows that this difference is significant from approximately 6 knowledge sources onwards. Panel C shows the average marginal effects of openness in innovation outcome, which is positively significant for firms with low levels of decentralization up to 6 knowledge sources and then turns negatively significant at about 8 knowledge sources, indicating an inverted 25
U-shape with a significant curvature. For firms with high levels of decentralization we find significantly positive marginal effects of openness until up to 7.5 knowledge sources, but the negative downward slope after the inflection does not turn significant. Panel D shows that the positive marginal effect of openness is significantly higher for firms with high decentralization compared to firms with low decentralization in the range of approximately 2 to 8 knowledge sources. Model 16 in Table 4, where decentralization is introduced via interaction terms, also provides evidence for a significant moderation effect of decentralization on the inverted U-shape relationship of openness on innovation outcome. All in all, we conceive this as support for Hypothesis 3 that decentralization positively moderates the effects of external knowledge search. On the basis of the empirical evidence of an upward shift as well as an increase in the inflection point in the relationship between openness and innovation outcome, we can say that high levels of internal decentralization are beneficial for firms such that they are able to gain more from a given (high) number of knowledge sources and to manage more knowledge sources.
------------------ Insert Table 3 about here --------------------
We now turn to the investigation of differences between firms with low versus high R&D intensity. Models 8 and 9 in Table 3 provide the baseline comparison without organizational variables. The results basically confirm the substitutional relationship between the use of external knowledge sources and internal R&D activities (Laursen
and Salter, 2006). Only firms with low R&D intensity seem to benefit from open innovation, albeit with diminishing marginal returns.2
------------------ Insert Figure 4 about here --------------------
Models 11 and 12 form the basis for investigating the effects of specialization in low and high R&D intensity firms. While the entire coefficient vector is significantly different (χ2(18)=57.11, p=0.00), the Wald test does show significant differences for neither the linear and squared terms of openness nor for the respective interaction terms. Except for the case of low R&D intensity and high specialization, Wald type tests find significant inflection points ranging from 6.5 to 7.6, but they are not significantly different. Panel A in Figure 4 displays the average prediction curves generated based on Model 11 and 12 for the four groups (low versus high R&D intensity and low versus high specialization, i.e. -/+ 2 standard deviations). It shows that in the case of low R&D intensity only firms with low specialization can benefit from open innovation up to a certain threshold of 6.5 knowledge sources, while firms with high specialization seem to lose from open innovation, albeit the downward slope is not significant. In the case of high R&D intensity, again only firms with low specialization can benefit from open innovation. They can do so even more effectively than their counterparts with low R&D intensity, as evidenced by differences in absolute predictions in Panel
This is also confirmed by average prediction and marginal effects which are not shown due to limited space. For firms with high R&D intensity, the marginal effect of openness remains insignificant over the entire range of openness. For low R&D intensity firms we find significant positive and negative marginal effects indicating the inverted U-shape. However, the absolute prediction of innovation performance of firms with low R&D intensity never reaches the performance level of high R&D intensive firms, not even at the inflection point.
B that start to turn significant from 5.5 knowledge source onwards. Furthermore, firms with high R&D intensity, but low specialization also do not experience a significant downward slope as their counterparts with low R&D intensity do. As evidenced by significantly negative marginal effects, firms with high R&D intensity and high specialization experience a performance decline if they go open, albeit this effect is somewhat neutralized again for extreme levels of openness. All in all, the effect of specialization works in the same direction for low and high R&D intensity, albeit on a higher performance level for R&D intensive firms.
------------------ Insert Table 4 about here --------------------
Models 14 and 15 form the basis for investigating the effects of formalization in low and high R&D intensity firms. The entire coefficient vector is significantly different (χ2(18)=60.41, p=0.00) in these two models. The Wald tests also show significant differences for the linear terms (diff=2.12, s.e.=0.78, p=0.01) and squared terms (diff=-0.15, s.e.=0.05, p=0.01) of openness, but not for the respective interaction terms. Wald type tests find significant inflection points only for low R&D intensity firms with low formalization (5.42 knowledge sources, s.e.=0.80, p=0.00) and with high formalization (8.30 knowledge sources, s.e.=0.75, p=0.00). This difference is significant (diff=2.88, s.e.=1.17, p=0.01) indicating an open innovation capability to manage more external knowledge sources for firms with high formalization, given low R&D intensity. Panel C in Figure 4 displays the average prediction curves which are generated based on Model 14 and 15 for the two groups low versus high R&D intensity as well as low versus high formalization; i.e. -/+ 2 standard deviations of the moderator 28
variable. It shows that in the case of high R&D intensity firms do not benefit from open innovation regardless of formalization, also revealed by insignificant marginal effects over the entire range of openness. In the case of low R&D intensity, more formalized firms can achieve higher performance by interacting with more knowledge sources. As Panel D shows, this difference is significant from about 7 knowledge sources onwards. All in all, we can conclude that the positive moderation effect of formalization that we found in the overall sample stems from the subsample of firms with low R&D intensity that may use open innovation in order to substitute for own R&D. Finally, models 17 and 18 form the basis for investigating the effects of decentralization in low and high R&D intensity firms. The entire coefficient vector is significantly different (χ2(18)=59.77, p=0.00) in these two models. The Wald tests also show significant differences for the linear terms (diff=2.33, s.e.=0.99, p=0.02) and squared terms (diff=-0.16, s.e.=0.06, p=0.02) of openness, as well as for the respective interaction terms (diff=1.45, s.e.=0.72, p=0.05; diff=-0.09, s.e.=0.05, p=0.05). Wald type tests find significant inflection points for all four curves in Panel E of figure 4, but only for the case of low R&D intensity and high decentralization the inflection point of 7.71 knowledge sources (s.e.=0.39, p=0.00) goes along with significant marginal effects. The other three curves do not seem to have a significant curvature as evidenced by insignificant marginal effects. Panel F, however, shows that firms with low R&D intensity do more extensively benefit form open innovation if they are decentralized rather than centralized. All in all, we can conclude that the positive moderation effect of decentralization that we found in the overall sample is more pronounced in the sample of firms with low R&D intensity, but is also shared to
small (albeit insignificant) extent with R&D intensive firms. That is why the positive moderation effect of decentralization in the overall sample is of larger effect size.
Engaging in multiple knowledge exchanges with external sources for innovation has become a "must" for many firms. Multiple studies and articles have postulated a new paradigm of open innovation. However, in our research we find that a more nuanced perspective on the readiness of firms to adopt open innovation is warranted. Using a sample of 365 German manufacturing firms, we confirm earlier studies that firms benefit from open innovation at diminishing marginal returns. Open innovation can even be too much of a good thing for some organizations, i.e. they engage in external knowledge search to a degree that is not compatible with internal capabilities and processes. With regard to firms’ open innovation readiness, the first important firm characteristic to look at is internal R&D intensity. We find that internal R&D is substitutional rather than complementary with external knowledge search, in line with Laursen and Salter (2006). As such, R&D intensive firms do not achieve additional gains from open innovation, while firms with low R&D intensity can successfully substitute own internal R&D, at least up to a certain degree of openness. To further explain the relationship between external search and innovation performance, we looked inside the firm and studied the effect of different organizational structures on open innovation performance. Scholars have repeatedly suggested that a firm’s organizational structure has a considerable impact on knowledge search and knowledge integration (Jansen et al., 2005; Li et al., 2008;
Rivkin and Siggelkow, 2003; Siggelkow and Levinthal, 2003). Yet, explicit investigations of the effects of a firm’s organizational structure on the generation of innovative knowledge via open innovation are scarce. Our study fills this gap by providing empirical evidence on the effect of a firm’s organizational structure on the performance contribution of external search strategies. We specifically examined the effects of specialization, formalization, and decentralization as main structural constructs determining an organization’s design. We find that specialization negatively moderates the effect of a broad search for external knowledge. Utilizing a large number of external knowledge sources translates into superior innovation performance only for those firms with low levels of specialization regarding the organization of their innovation activities. On the contrary, firms who focus their innovation activities in a specialized organizational unit seem not to profit from open innovation. Similarly, we find that firms that have structured their innovation activities according to a product, technology, or region are much less likely to turning external search into performance than firms that have a broader, less structured organization. The negative moderating effect of specialization especially holds for R&D intensive firms, which we found to have difficulties in profiting from open innovation on average. However, this result also has an important implication for R&D intensive firms. If they are able to implement a less specialized organization of their innovation activities, they may be also able to realize a complementary alignment between their internal R&D and external knowledge search, and ultimately a performance gain by open innovation. This result qualifies the finding by Laursen and Salter (2006) who proposed complementary relationship between external search and internal R&D based on an absorptive capacity logic, but find a substitutional one. Post hoc, they
explain this finding by the “not-invented-here (NIH)” syndrome that may increase with internal R&D intensity. Now, we can somewhat explain this finding by the degree of specialization of innovation activities. The absorptive capacity argument regains explanatory power over the “NIH” argument in firms that manage to spread their R&D and innovation budget over the broader organization rather than financing specialist departments that may be more heavily affected by “NIH” attitudes. Our findings also challenge current research that has associated specialization with positive effects due to higher propensities to search outside (Olson et al., 2005; Rothaermel and Hess, 2007). One could have assumed that specialized organizations have a better capacity to utilize external knowledge and to exploit it in the later stages of innovation processes (Damanpour, 1991). But our results confirm an opposite view that specialization is generating self-focused silos of knowledge generation and exploitation. Specialists were found to engage in “narrow” search endeavors (Van den Bosch et al, 2003) and to create mental and structural boundaries (Olson et al., 2005; Pertusa-Ortega et al., 2010). With increasing interunit boundaries, the costs of communication and learning rise, and knowledge transfer is hampered (Colombo and Delmastro, 2008; Willem and Buelens, 2006). Profiting from open innovation demands "shared mental representations" (Knudsen and Srikanth, 2010) which intuitively are developed better in a non-specialized organization. This means, however, not that firms should never specialize their innovation activities. However, our results suggest that these firms have to build a dedicated selection capability to identify the (very) few external sources which they then utilize to a high amount. In such a situation, firms could benefit from Rothaermel and Hess’ (2007) “star scientists” or “gate-keepers”, who manage the interface between a firm and its external environment (Tushman, 1977).
Formalization positively moderates the effects of external knowledge search on innovation performance. Our research shows that the ability to benefit from open innovation and external knowledge search is strongly supported by formalized procedures and dedicated planning tools. While we could not find that firms with formalized innovation processes are able to manage more knowledge sources, we find that they benefit significantly more from a given number of external knowledge sources than firms with low formalization. Our results suggest that defining clear rules, mechanisms, and responsibilities for conducting the search and transfer may alleviate the information overload of hitherto unspecified knowledge search and transfer procedures. Formalized processes also seem to define procedures to evaluate external knowledge (cf. Pertusa-Ortega et al., 2010). Decentralization, i.e. the delegation of authority for innovation to lower levels in the hierarchy, enhanced by dedicated training and HR tools, positively moderate the performance effects of open innovation and external knowledge search. The organizational structures and supporting measures that correspond to a high degree of decentralization enable firms to gain more from a higher number of knowledge sources. In addition, decentralization shifts the inflection point of the inverted Ushape of the openness-performance curve almost to its (right) extreme, implying that also more knowledge sources can successfully be managed. The positive moderation effects of both higher formalization and decentralization are even more pronounced for firms that have a low internal R&D intensity. These results have an important implication for firms that try to substitute their low internal R&D intensity by the means of open innovation. For them, higher formalization and decentralization can enhance the desired substitutional effect.
Building on a famous quote by Drucker (1985), today no one needs to be convinced any longer of the importance of open innovation – how to openly innovate is the key question. Our research contributes to this investigation, addressing recent calls to study the governance of open innovation initiatives in larger detail. Taking a contingency perspective, we focus on the organizational structures that facilitate open innovation. We contribute to the literature by a more immediate and finegrained theoretical connection between the utilization of external sources of knowledge and organizational structur