Modes of Innovation: A Grounded Meta-Analysis

This study takes a meta-analysis drawing on components of grounded theory approach to provide a more comprehensive analysis and synthesis of studies on modes of innovation, in particular, on the science, technology, and innovation (STI) and doing, using, and interacting (DUI) modes. This research provides a meta-analytical review of research purposes, methodologies and investigates the existing inconsistencies in findings of relevant qualitative and quantitative studies from 2007 to 2014. The majority of studies aim at investigating modes of learning and innovation and identifying the most effective mode. In addition, firms in different country contexts, which combine the STI and DUI forms of learning, are more likely to innovate. The implications of meta-analysis are intended to benefit policy makers, researchers, and practitioners.

, or on the basis of high and low-technology are problematic. During times of dramatic changes in technologies and production, such distinctions tend to become blurred. Jensen et al. (2007) introduced a taxonomy, which distinguishes between the science, technology, and innovation (STI) and doing, using, and interacting (DUI) modes of innovation. The first mode emphasizes the scientific and technologybased nature of innovation. The second mode is based on experience and interactive practice. The third mode is a combination of the former two modes. The seminal study of Jensen et al. (2007) bridges the gap between the linear approach to innovation and the systems, which entails interactions (Leydesdorff 2012) and practice.
Recently, the debate on the most effective mode of innovation and types of collaboration has attracted interest among scholars around the world (Jensen et al. 2007;Aslesen et al. 2012;Parrilli and Elola 2012;Fitjar and Rodriguez-Pose 2013;Gonzalez-Pernia et al. 2014). Even within the same theoretical framework, studies can vary in methodology design, data collection method and, consequently, the findings are contradicting. This study aims to provide a more comprehensive analysis and synthesis of studies on the STI and DUI modes of innovation and collaboration within these modes. We conducted a grounded meta-analysis (Glaser and Strauss 1967;Strauss and Corbin 1994;Hossler and Scalese-Love 1989) that enables to combine both quantitative and qualitative studies, extract comparable categories from the studies, and compare studies from standpoints of research purposes, methods, and results. This paper provides a grounded meta-analysis of the existing studies on the STI and DUI modes of innovation. A major purpose of this type of analysis is to provide a comprehensive comparative analysis of studies on business modes of innovation, and to inform researchers, policy makers, and practitioners about the state-of-the-art and future research directions in their area of interest. This analysis encompasses 17 studies on the STI/DUI modes of innovation published from 2007 to 2014. We synthesize the research purposes, methodologies, and results of the studies that help to identify trends and patterns in this research field.
The paper is structured as follows. In section 2, we review the literature on STI/DUI modes of innovation and provide a comparative discussion of these modes and studies. Section 3 describes the methodology of grounded meta-analysis together with the selected variables and data as well as their coding procedure. In section 4, the results of grounded meta-analysis are provided. In the final section 5, the study and findings are summarized and discussed together with the relevant implications for researchers, policy makers, and managers. In addition, we point out the limitations and issues that should receive attention and delineate future research lines.

Introducing the STI/DUI Modes of Innovation
The STI mode emphasizes that innovation is based on research and development (R&D), human capital (scientifically trained personnel), and research collaborations. This mode relies mainly on science-based knowledge and further develops explicit, global know-why, and know-what type of knowledge (Jensen et al. 2007). Human capital involves employees with a PhD (masters) in natural sciences or in construction engineering (Jensen et al. 2007) that are involved in innovation projects. A high level of scientific education across the employees increases their absorptive capacity and, consequently, improves the impact of R&D activities (Cohen and Levinthal 1989). Firms benefit in terms of innovation capability from a stronger connection to science that provides a platform for firms' technological innovation and learning. The main STI partners are researchers, universities, and research organizations (Isaksen and Karlsen 2012b;Jensen et al. 2007;Parrilli and Elola 2012;Fitjar and Rodriguez-Pose 2013).
The DUI mode of innovation emphasizes innovation based on learning-by-doing, by-using, and by-interacting. The concept of learning-by-doing (Arrow 1962) implies that a firm performs experiential learning and increases productivity and efficiency by getting more practice and repeating the same operations. Learning-by-using relevant state-of-the-art technologies helps to acquire competences to increase the productivity of machines (Rosenberg 1982) and learning-by-using user experience creates opportunities for experimentation and problem-solving on the shop floor (Wuyts et al. 2004;Lorenz 2012). Innovation can also be a result of interactions, networks, informal relationships, and organizational collaborations within and between organizations (Audretsch 2003;Lundvall 1992;Gemünden et al. 1996;Fu et al. 2013). DUI interaction implies close cooperation with customers, suppliers, distributors, and competitors (Jensen et al. 2007;Fitjar and Rodriguez-Pose 2013;Gonzalez-Pernia et al. 2014). At the level of a firm, the DUI mode can be characterized as decentralized decision-making, softened hierarchies, eliminating strict boundaries between functions, and intensive teamwork. This mode is defined as a user-driven mode that supports the development of new products and services in compliance with market needs (Isaksen and Nilsson 2013). Table 1 provides a brief synthesis of the modes of innovation based on analyzed studies: Jensen et al. 2007;Sanchez 2008;Chen, et al. 2011;Aslesen, et al. 2012;Isaksen and Nilsson 2013;Isaksen and Karlsen 2010;Parrilli and Elola 2012. Since the dominant knowledge base in the DUI mode of innovation is tacit (Jensen et al. 2007), the knowledge acquisition and exploitation require frequent interactions with external business partners (Lundvall 1988;Nonaka and Takeuchi 1995), which can be facilitated by different kinds of proximities (Presutti et al. 2011). DUI firms collaborate externally with customers, suppliers, distributors, and competitors, while the main partners of STI firms are researchers, universities, and research organizations. The STI mode prevails in research-oriented industries such as pharmaceuticals and chemicals, aeronautics, petroleum, and nanotechnology. In contrast, the DUI mode dominates in more traditional industries, for example, traditional manufacturing. The STI mode aims at developing more radical innovations (product, process) (Jensen et al. 2007). However, the DUI mode may also facilitate the generation of radical innovations (Lorenz 2012), for example when organizational innovation is involved (Apanasovich 2014). The threat of knowledge leakage is higher in the STI mode of business innovation because of the use of codified knowledge, which is not sticky and can be more easily transferred to the recipients that count with a proper absorptive capacity (Sanchez 2008;Chen et al. 2011). Therefore, managers decide to patent some inventive products (Acs et al. 2002) and in other occasions intentionally keep some knowledge in tacit form in order to prevent the flow of knowledge to competitors (Schulz and Jobe 2001).
Sometimes firms apply the "pure" STI or "pure" DUI mode of innovation, in other occasions they tend to combine the two modes. A third mode is a combination of the two modes. This mode combines different innovation drivers (science and technological drivers with learning-by-doing, using, and interacting). Jensen et al. (2007) argue that firms combining the STI and DUI mode of learning and innovation are more likely to innovate than those using the STI or DUI alone. The ongoing contention on identifying the most effective mode of innovation has attracted the attention of international scholars (Aslesen et al. 2012;Isaksen and Nilsson 2013).

Grounded Meta-Analysis
The quantitative meta-analysis is an application of different statistical methods to collect, combine, contrast, and identify patterns among different studies that focus on the same topic. On the one hand, it enables to extract and aggregate empirical findings from multiple quantitative studies and transform into a common measure (Rosenbusch et al 2011). With the help of statistical procedures, the results can be compared and evaluated (Stanley 2001). This type of analysis provides research integration, comparison, and generalizability of results and interpretations (Hunter and Schmidt 2004). The use of quantitative techniques is appropriate when the aim of the analysis is to identify statistically justified relationships between variables and when there is a relevant amount of quantitative studies (Hossler and Scalese-Love 1989). The limitation of this type of analysis is that it excludes qualitative information (Stall-Meadows and Hyle 2010), which is important in our field of study.
To conduct our research, we were looking for a method that would allow us to synthesize both quantitative and qualitative inquiries and to compare studies from standpoints of research purposes, methods, and results. Qualitative meta-analysis using a grounded method (grounded meta-analysis) (Glaser and Strauss 1967;Strauss and Corbin 1994;Hossler and Scalese-Love 1989) is a reasonable alternative as it enables combining both quantitative and qualitative studies, extract comparable categories from the studies, and overcome the aforementioned limitations.

Methodology of Grounded Theory
We conducted meta-analysis based on the grounded theory approach. The specific methodology of grounded theory was discovered and developed by Glaser and Strauss, mostly through the analysis of interviews and observations (Glaser and Strauss 1967;Strauss and Corbin 1990;Corbin and Strauss 1990). The grounded meta-analysis of published studies was firstly introduced by Hossler and Scalese-Love (1989) and then conducted and developed by other scholars (Ke 2009;Wu et al. 2012;Stall-Meadows and Hyle 2010). This method was chosen because it enables us to synthesize the theory, explore methods and review empirical findings, and to formulate a complete image of this field of study (Hossler and Scalese-Love 1989). Moreover, the grounded meta-analysis enables us to combine both qualitative and quantitative studies that are essential for an exhaustive synthesis of the literature (Rahimi et al. 2009;Stall-Meadows and Hyle 2010;Timulak 2009). The data have been extracted from 17 studies on the STI/DUI modes of innovation conducted in Europe, Asia, and North America. This study aims to investigate the modes of innovation and their impacts on business innovation performance in different economic systems. To reach this goal, we review studies published since 2007 with the explicit focus on STI/DUI modes of innovation.

Data Collection
To select appropriate studies for the analysis, we have considered the following inclusion criteria: (i) content devoted to the STI/DUI mode of innovation, (ii) year of publication (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014), and (iii) English language publications. In order to identify relevant studies, a systematic search was carried out in the data pool consisting of electronic databases (e.g., EBSCO, Scopus, Emerald, Elsevier Science Direct). Special attention was paid to identify journals where the most influential articles were published. Then, manual searches were conducted among the selected journals issued. As there was lack of studies in peerreviewed journals, we also conducted a search through working papers series (e.g., Orkestra, Lund University), as well as in conference proceedings (e.g., International Schumpeter Society Conference), and dissertations (e.g., University of Lisbon). Additionally, we reviewed relevant works in their reference sections so as not to exclude any other meaningful study. Then, we tried to identify scholars who specialize in this field of research to conduct additional searches by authors' names. When we did not have an access to the full version of papers, we contacted the authors personally. As a result, our final sample comprised 17 studies on modes of innovation.

Data Coding and Analysis
We coded quantitative and qualitative information in compliance with the coding procedure set by Hossler and Scalese-Love (1989) which is widely used in grounded meta-analysis (Ke 2009;Wu et al. 2012;Stall-Meadows and Hyle 2010). While the coding of quantitative data requires beforehand logically deduced categories, the coding of qualitative data requires inductively identifying the categories after carefully reviewing the data (Hossler and Scalese-Love 1989). The coding is the interpretive process, which enables constant comparison of data and revision of coding categories (Glaser and Strauss 1967;Hossler and Scalese-Love 1989;Strauss and Corbin 1990;Stall-Meadows 1998). After a thorough perusal, we developed a coding matrix to delineate the studies by their research purposes, methods, sample sizes, types of firms, countries, indicators of modes of innovation, and empirical findings. A summary of these studies is presented in Table 2.

Results of Grounded Meta-Analysis
The grounded meta-analysis is more a deductive than inductive process and requires the investigator to start their analysis without clearly determined set of research questions or hypotheses (Hossler and Scalese-Love 1989). Specifically, our study addresses the following questions: 1. What are the major research purposes of studies on modes of innovation? 2. What are the major methodologies of studies on modes innovation? 3. What are the most appropriate indicators of the STI and DUI modes of innovation? 4. What is the most effective mode of innovation across countries? Both STI and DUI interaction matter for innovation. Within the DUI mode, collaboration with extra-regional agents is much more conducive to innovation than collaboration We categorized each article into two groups according to its research purpose: #1 it investigates modes of learning and innovation and identifies the most effective one, and #2 it explores the sources of firms' knowledge and the types of interaction. The most frequent is the research purpose #2 to investigate the modes of learning and innovation (65 % of analyzed studies). For the rest of the articles, the main purpose is #1 to explore the sources of firms' knowledge and the types of interaction.

Research Purpose 1: Exploring STI-DUI-Firm Interaction
The studies pertaining to the first group according to their research purpose explore diversity in interaction and knowledge sources of small and medium enterprises (SMEs)-relying on different modes of innovation.  (2010) showed that universities often cooperate with industries and businesses dominated by the STI mode of innovation (e.g., biotechnology industry). However, cooperation between universities and industries dominated by the DUI mode of innovation also takes place mostly in the form of education and training of personnel. The empirical analysis conducted by Chen et al. (2011) illustrates that firms relying on the DUI mode benefit from relationships with value chain partners and competitors as well as with technology-related organizations. In addition, they found that firms relying on the STI mode profit from collaborations with universities and research institutes, as well as with value chain partners. The analysis conducted by Trippl (2011) is based on the case study of 10 firms in the Vienna food sector and 10 interviews with research organizations and industry experts. It was found that the DUI collaboration mainly takes place with extra-regional partners, while STI proved to be mainly regional. Different results were found by Isaksen and Karlsen (2012a). According to this study, firms relying on the STI mode source most of their knowledge outside the region, while firms relying on the DUI mode of innovation tend to focus their knowledge sourcing in the region.
Fitjar and Rodriguez-Pose (2013) investigated two types of firm interactions (STI/ DUI) based on a sample of 1604 firms in Norwegian city regions. By means of a logit regression analysis, they test the relationship between the STI and DUI modes separately with product and process innovation. Attention is also paid to the geographical location of partners. The results of regression analysis show that the collaboration with extra-regional partners is much more relevant to innovation than collaboration with local agents. Fitjar and Rodriguez-Pose (2013) have revealed that different types of partnerships are related to different types of innovation. From the partnerships linked to the STI mode of innovation, only interactions with universities (but not with research institutes and consultancies) are relatively strongly associated with radical product and process innovation. DUI cooperation with customers and suppliers is closely related to the innovative capacity of firms, while cooperation with competitors has a detrimental effect on innovation. Collaboration with suppliers has a strong positive impact on both types of innovation, especially on process innovation. Collaboration with customers is closely related to product innovation, however, does not significantly affect process innovation. Gonzalez-Pernia et al. (2014) analyzed the relationship between DUI and STI collaboration and firm innovation (product and process). The authors conducted a regression analysis on the sample of 4969 firms over an average period of 7.1 years. The results of analysis show that firms that combine DUI and STI types of partnerships are more effective in generating product innovation, and process innovation benefits more from the DUI-related partnerships undertaken by firms.

Research Purpose 2: Identifying the Most Effective Mode of Learning and Innovation
The most popular research purpose across the analyzed studies is to identify the most effective mode of innovation. In their pioneering study, Jensen et al. (2007) show that firms in Denmark that use the mixed STI and DUI modes of innovation are more innovative than the rest. In their study, the modes of innovation are tested using data from the 2001 Danish DISKO survey. On the sample of 692 firms, they perform a cluster analysis to identify groups of firms that practice different modes: low learning, STI, DUI, and DUI+STI clusters. However, Fitjar and Rodriguez-Pose (2013: 2) consider that such division "implies a significant loss of information about STI and DUI modes of learning at the level of each firm" because of the strict assignment of the firm to one of these four categories. The innovation output in the article by Jensen et al. (2007) is measured in terms of product and service innovation. A classification of this variable is the following: no innovation, new-to-firm, new-to-national market, and newto-international market. The authors show that the combination of the STI mode of innovation and the DUI mode is strongly correlated with innovation output.
Later, Chen and Guo (2010), Aslesen et al. (2012), Isaksen and Nilsson (2013), Isaksen and Karlsen (2012a), Nunes et al. (2013), Amara et al. (2008), and Apanasovich (2014) confirm totally or in part the results produced by Jensen et al. (2007). The study by Chen and Guo (2010) is based on the sample of 230 Chinese manufacturing firms. According to the authors, firms combining the STI and DUI modes of innovation are more likely to innovate than those emphasizing the STI or DUI mode separately. Aslesen et al. (2012) analyze the dominant modes of learning and innovation based on a sample of 96 firms in the Agder region in Norway. This study corroborates the result obtained by Jensen et al. (2007); the combination of the two modes of innovation is the most effective business strategy. According to Aslesen et al. (2012), firms that use the STI+DUI mode of innovation respond differently to the challenges of globalization. For example, the DUI firms face a high risk of failure because of high competition from low-cost countries. The STI firms, in turn, face the threat of being relocated to another country if they do not ensure a strong competitive advantage to their international or national business groups. Nevertheless, firms that combine the STI and DUI mode of innovation are more likely to ensure higher competitiveness thanks to their strong regional technological base. On the sample of 639 technology manufacturing Canadian SMEs, Amara et al. (2008) test the relationship between learning and novelty of innovation of the established SMEs by means of regression analysis. They deduce that learning by searching (investment in R&D, scientific trained personal, STI mode) and learning by training, doing, using, and interacting (DUI) impact positively on the degree of novelty of innovation in SMEs.
In contrast to the abovementioned studies, on the basis of empirical evidence from Spain, Parrilli and Elola (2012) show that the product innovation is, in fact, more sensitive to the STI drivers than to the DUI drivers. Three modes are tested on a sample of 409 Spanish SMEs in the Basque country (region in Spain) mostly belonging to medium-technology manufacturing and high-tech knowledge-intensive services industries. An ordinal regression analysis is performed to evaluate the impact of modes of innovation on product innovation. In their case, the combination of STI+DUI innovation modes does not add value (i.e., innovation output) to the adoption of the STI mode alone. In the study based on data from Spain, Gonzalez-Pernia et al. (2012) demonstrate that the combined effect of the STI+DUI mode does not seem to improve the effect on innovation output (i.e., product and process) vis-a-vis the separate effect of the STI and DUI modes. In contrast to the previous studies that are based on one time period, Gonzalez-Pernia et al. (2012) apply a multi-period approach by using large longitudinal data. In order to verify the relationships between the innovation drivers and their output, panel regression tests were conducted using 33,789 observations from 8,500 Spanish firms. A recent study by Apanasovich (2014) concludes that the SMEs in Belarus that combine the STI and DUI modes of innovation are more likely to generate product innovation. However, SMEs that rely on the DUI mode alone are more likely to generate product innovation than those that rely on the STI mode alone. In contrast to product innovation, firms combining the STI and DUI modes of innovation are not more effective in generating organizational innovation than firms relying on the DUI mode alone.

Research Question 2: Distribution of Research Methodologies
Analyzed studies were categorized into two groups according to their research purposes as follows: #1 to investigate modes of learning and innovation and identify the most effective one, and #2 to explore the sources of firms' knowledge and the types of interaction.The majority of studies with purpose #1 are quantitative (70 %). However, studies with research purpose #2 are equally distributed between research methodologies. In the majority of quantitative studies, regression analyses were conducted, while qualitative studies are almost completely based on case study methodology.

Research Question 3: Indicators of the STI and DUI Modes
Researchers empirically assess different typologies of modes of innovation with the help of the indicators. There is a big diversity of indicators of the DUI mode, while indicators of the STI are more common. Notwithstanding the lack of standard indicators, we extracted 3 indicators that define the STI mode and 13 indicators that characterize the DUI mode of innovation (Table 3). We have not included exact measures of indicators because of the wide diversity of scales (binary, ordinal, and absolute values). The first group of STI indicators emphasizes that innovation is the result of science and R&D (Jensen et al. 2007;Parrilli and Elola 2012) suggesting that investment in R&D and scientific human capital are considered as key innovation inputs. These indicators are more common in the analyzed studies, thus, have not evolved much since they were introduced by Jensen et al. (2007). In this regard, the first indicator of the STI mode is the expenditure on R&D (mostly as a percentage of the annual turnover or in absolute terms). The next indicator is the number of R&D personnel. The third indicator assesses the level of STI interaction, i.e., cooperation with researchers attached to universities or research organizations. Unlike the STI mode, the DUI mode stresses the practical and interactive nature of innovation. This mode of innovation stresses that the innovation is based on learning-by-doing, byusing, and by-interacting (Arrow 1962;Rosenberg 1982;Lundvall 1992;Jensen et al. 2007). The indicators of the DUI mode are diverse and heterogeneous across the analyzed studies. The indicators introduced by Jensen et al. (2007) are capable of assessing only interactive (I) aspects of the DUI mode. As the methodology of measuring the DUI mode evolved over time, new indicators were proposed. Apanasovich (2014) has grouped the indicators in three categories on the bases of three aspects of the DUI mode: doing (D), using (U), and interacting (I). The author firstly proposed the first two drivers such as preliminary marketing and technological preparation for production to assess the "D" (doing) and "U" (using) aspects of the DUI mode. The other 12 indicators measure "I" (interactive) aspects or internal and external business relationships established by the firm. Internal DUI interaction works within a firm top-down and bottom-up (vertical communication) and between different company departments (horizontal communication) (Hinds and Kiesler 1995). The indicators of DUI internal interaction are interdisciplinary workgroups, quality circles, systems for collecting proposals, and autonomous groups. Some indicators (integration of function, softened demarcations) measure whether a firm has a flexible and decentralized organizational structure. The indicators assessing external DUI interaction show whether a firm collaborates in innovation activities with customers, suppliers and distributors, competitors, and other organizations (Fitjar and Rodriguez-Pose 2013;Fu et al. 2013). All indicators are depicted in Table 3.

Research Question 4: The Most Effective Mode of Innovation
Via coding procedures (Hossler and Scalese-Love 1989;Stall-Meadows and Hyle 2010;Corbin and Strauss 2007), we synthesize the data about the most effective business innovation mode (Table 4). Most in-country analyses focuse on countries that operate in market economies: Denmark (Jensen et al. 2007), Norway, Sweden (Aslesen et al. 2012;Isaksen and Nilsson 2013;Karlsen 2012a), Portugal Nunes et al., (2013), Canada (Lorenz 2012), and Spain (Parrilli and Elola 2012). In contrast, only one that has appeared recently is based on firms' data from countries that operate in transition economies (Apanasovich 2014).
We can observe that for almost all analyzed studies with the research purpose (№2) (investigate modes of learning and innovation and identify the most effective one), the STI+DUI mode is identified as the most effective one. Thus, firms in Denmark, Norway, Sweden, Portugal, and China perform more effectively through the combination of the STI and DUI modes of innovation. According to the analysis of SMEs in post-soviet countries in transition, firms combining the STI and DUI modes of innovation are more likely to generate product innovation (Apanasovich 2014), while firms relying on DUI mode alone are more likely to generate organizational innovation. In contrast, Parrilli and Elola (2012) show that product innovation is more sensitive to STI drivers than to DUI drivers. The authors tested the modes of innovation on a sample of Spanish SMEs (in the Basque region) that belong to medium-technology manufacturing and high-technology knowledge-intensive service industries situated in  (Parrilli and Elola 2012). On the basis of these different empirical data and studies, we conclude that, regardless of the country's geographical and economic context, the most effective mode of innovation is the STI+DUI. However, some deviations can be observed in studies based on one or several regions or types of chosen industries (low-or high-technology manufacturing industries, high-tech or less knowledge-intensive services, etc.).

Conclusion
This paper is designated as grounded meta-analysis of studies of STI/DUI modes of innovation. This meta-analytic approach gives the opportunity to analyze studies that focus on the same topic, but apply different methodological designs. Categories such as research purpose, methodology, indicators, and findings have been extracted. Considering the limited number of studies and rather new theoretical framework of the STI/DUI modes (starting from 2007), we realize that our results should not be totally generalized. However, the results of our study highlight several issues that could help researchers to focus on studying the phenomenon, policy makers to promote programs that support the innovation mode of firms (industry), and help managers to foster effective innovation performance in their companies. This study is a relevant reference basis for future research on a firm's modes of innovation. The meta-analysis can be useful for young researchers as it provides a comprehensive analysis of the state of the art in their area of interest.
The attention of international researchers has been paid to the ongoing debate on the most effective mode of SMEs innovation (Chen, et al. 2011;Aslesen, et al. 2012;Parrilli and Elola 2012;Isaksen and Nilsson 2013). The studies have shown that firms in Denmark, Norway, Sweden, Portugal, Canada, China, and Belarus combining the STI and DUI modes of innovation are more likely to innovate than those relying on the STI or DUI mode alone.
For further improvements, we would like to highlight the following issues. First, researchers can face difficulties in assessing empirically typologies of modes of innovation because of the large diversity and lack of standard indicators of the STI and DUI modes (especially the DUI mode). The future research line could be to further identify reliable indicators, which are essential for high-quality empirical analyses.
The second issue relates to a possible problem of reverse causation (endogeneity) between the dependent variable and independent variables in quantitative studies. This problem must be addressed by including in the model the value of explanatory variables with a lag (Bilbao-Osorio and Rodriguez-Pose 2004; Audretsch and Keilbach 2004) compared to the dependent variable. Moreover, it can take several years until the R&D activities and interaction (Rosenberg 1990) can start generating a cash flow and affect industrial productivity. The majority of studies use only data from a 1-year period. Longitudinal data would be more relevant to verify such tendencies over a longer time span.
However, studies of modes of innovation do not show what the composition of innovation drivers is in the combination of the STI and DUI mode of innovation. In this regard, future research could explore how different drivers of innovation could be combined within the firm.