Determinants of scientific production: An empirical study of the world s top R&D companies. Michele Cincera and David Dratwa

Transcription

1 Determinants of scientific production: An empirical study of the world s top R&D companies Michele Cincera and David Dratwa Solvay Brussels School of Economics and Management, Université Libre de Bruxelles Work in progress: September 2011 Abstract This paper aims at assessing the determinants of firms scientific production as measured by the publication of scientific papers. As stressed by some previous studies, firms from the private industry tend to publish extensively in the scientific literature. However, up to now few empirical studies have been conducted to assess the determinants of firms scientific production. The paper begins with a review of the motives for firms to publish their knowledge in scientific journals. Descriptive statistics provides trends in the output of corporate scientific research. The following empirical analysis covers the 2,000 largest R&D companies over the recent period ( ). The results show that the size of the firm s R&D as well as collaboration matters for the quantity and the quality of the scientific knowledge produced by the firm. 1

2 Introduction The traditional view seeing science and technology as distinct entities appears less convincing as interactions between those spheres tend to intensify. In this respect, an interesting matter is the contribution of industry research to the development of open scientific knowledge. Companies belonging to the private industry contribute to the advancement of publicly available knowledge by increasingly publishing the results of their research through scientific papers (Hicks, 1995; Stephan, 1996; Godin, 1996). At first sight, this behaviour may seem puzzling on behalf of organizations driven by logic of profits and seeking the appropriation of the benefits resulting from their research efforts. As we will show in this paper, various reasons may explain why firms reveal their knowledge by publishing scientific papers. Besides a few exceptions (Stern, 2004; Adams 2010; Simeth and Raffo, 2011), scarce are the empirical studies on the determinants of firms scientific production. In this paper, we will try to provide further evidence on the determinants of firms scientific production as measured by the publication of scientific papers. More specifically, this paper aims at assessing the determinants of scientific publication s activity by the largest R&D companies over the recent period ( ). Among these determinants we consider the firms own R&D; the size of the firm; the share of co-authorships with national firms; foreign firms; national universities; foreign universities; other national public research institutions and other foreign public research institutions; the number of citations received. As control variables we take into account the distribution of the firm s publications across industry sectors where the firm operates as well as country and year dummies. The dataset used in the empirical analysis consist of two main data sources. The first data source is the last edition of the EU industrial R&D scoreboard released in 2010 by the JRC- IPTS of the European Commission. The R&D scoreboard has been issued every year since 2004 and provides data at the firm level for the top 1,000 R&D-active firms in the EU-27 and the top 1,000 outside the EU-27. The Elsevier s Scopus online database is the second main data source. The information concerning the scientific publications of each of the 2,000 top R&D firms have been retrieved manually from this database. The paper presents several descriptive statistics which illustrates by means of a series of figures the main trends and facts about scientific publication activity of the world s largest 2

3 R&D companies. The empirical findings are obtained by means of several econometric models for count data in the context of panel data. These models are implemented to estimate the impact of the various determinants of firms scientific production. The remainder of the paper is organized as follow. Section 2 provides a literature review about industry scientific production and the motives for firms to publish their research in scientific journals. Section 3 describes data sources and sample composition. Section 4 provides descriptive statistics on publication trends of the world s top R&D companies at the country and sector level. Section 5 presents and discusses the results from the econometric analysis. A final section provides conclusion. 2. Literature review In this paper, the focus will be put on the output side of the production of knowledge and more specifically on scientific papers as outcome of industry scientific research. The first and foremost condition to publish scientific knowledge in papers is the production of this knowledge. The production of scientific knowledge is inherent to the conduction of and investment in research activities and more precisely fundamental or basic research. What leads firms from the private industry to undertake research and to produce scientific knowledge? To address this question let us recall Rosenberg (1990) who provided insights of the reasons for firms to invest in basic research. Rosenberg argued that: (1) basic research is a long-term investment (2) that gives first-mover advantages that can be used as barrier to the entry of new firms (3) the outcome of basic research is often unpredictable and generated unintentionally (4) basic research is a ticket of admission to an information network (5) firms often need to do basic research in order to understand better how and where to conduct research of a more applied nature and is essential to evaluate the outcome of much applied research and for perceiving its possible implications (6) basic research offers a valuable instrument to monitor and to evaluate research being conducted elsewhere. Rosenberg s latest argument ties up with the view of Cohen and Levinthal (1989, 1990) that firms invest in research not only to generate new knowledge but also to build an absorptive capacity i.e. to enhance their ability to identify, assimilate and exploit externally generated knowledge. This emphasis on the learning role of R&D suggests that firms may thus not benefit from the knowledge produced elsewhere unless they perform themselves 3

4 research. Following Cohen and Levinthal, a firm s absorptive capacity is build through the development of a stock of prior knowledge (i.e. the firm s knowledge base) that R&D contributes to increase. Some recent empirical studies focused on the determinants of the production of industrial scientific knowledge at the industry and firm level. Adams and Clemmons (2006) investigated the role of industry basic research and R&D spillovers in the production of scientific knowledge at the industry level. Using papers as scientific output, they provided evidence that industry R&D stock as well as academic and industrial spillovers have a significant positive effect on the production of scientific knowledge. By measuring the effect by the elasticity of scientific papers with respect to basic research flows, they found, among other results, that the academic spillover elasticity exceeds the industrial spillover elasticity and the elasticity of the industry R&D stock. In a related work, Adams and Clemmons (2010) estimated a production function for industrial science at the firm level. They considered the firm s basic research stock as well as outside knowledge stock (i.e. university and other firms past scientific research) as determinants of the firm s scientific production. They provided evidence that outside knowledge matters more than inside knowledge in the firm s scientific production, and that knowledge from universities matters more than knowledge from other firms. Finally, Adams and Clemmons found that the role of outside knowledge in the production of industrial science increases when aggregating from the firm to the industry level. After the scientific knowledge has been produced, the firm has to decide whether to disclose it (i.e. through patents or publications) or not (i.e through secrecy). The following section will explore the reasons that could lead firms to disclose their knowledge by publishing papers in the scientific literature. Publications in scientific journals have the characteristics of an open knowledge disclosure or a voluntary spillover. This has three implications. The first one is that the knowledge is intentionally disclosed that is to say that the disclosing firm has the choice to keep it secret if they want it. The second one is that the access to the disclosed knowledge is free of charge. The disclosing firm is thus not remunerated or at least not directly as we will try to show in this section. The third one is that the access to the knowledge is open, which means everybody can benefit from it. Once disclosed in scientific papers, the knowledge becomes a public good. Publication in scientific journals is not the only way through which firms may openly disclose their knowledge. For instance, it is common for firms to reveal 4

5 knowledge through conference presentations (Pénin, 2007). Furthermore, the Internet can be seen as another channel through which knowledge can be disclosed, especially in the case of the development of open source software (von Hippel and von Krogh, 2003). At first glance, it seems puzzling why profit-seeking firms would voluntarily disclose knowledge through publications given their objective to generate returns from their R&D. Indeed, open knowledge disclosure entails costs for the firm, the most important being the communication of information to existing and potential competitors involving the loss of competitive advantage. Moreover, the firm is not directly remunerated for revealing its knowledge. This section will investigate the motivations for profit-seeking firms to openly disclose their knowledge in scientific publications and hence to contribute to publicly available knowledge. We will try to show that firms may in fact reap indirect benefits from adopting such practice. 2.1.Knowledge Transfer In Vertical Interactions Practice of open knowledge disclosure can occur in the context of vertical knowledge transfer between suppliers (or manufacturers) and buyers (or users). On the supplier side, openly revealing knowledge may have two purposes: increasing production in downstream sectors and triggering feedback from users. While on the user side, open knowledge disclosure is motivated by cost reduction and productivity increase. a. Supplier side Suppliers may be motivated to publish useful information in order to increase production in the downstream sector and hence to increase demand of intermediary goods that is addressed to them (Harhoff, 1996). Harhoff developed a two-industry model consisting of a monopolist supplier and oligopolistic buyers that are vertically related. Harhoff noted that the monopolist supplier could foster product improvements in the downstream industry by generating knowledge spillovers that can substitute for downstream firms own R&D efforts. The upstream supplier may benefits from disclosing knowledge to downstream firms through two effects. First, knowledge disclosure decreases the sunk costs of R&D for downstream firms and thus reduces entry barriers at the downstream level. This will in turn ease the entry of new firms and thus increase the number of competitors in the downstream industry. 5

6 Second, if the structure of the downstream industry is exogenous 1, disclosure allows the supplier to affect downstream product quality by increasing the level and productivity of the research in the downstream sector. This in turn decreases production costs and increases overall production at the downstream level. Consequently, those two effects lead to an expansion of downstream output and increase the downstream firms demand for intermediate goods provided by the supplier. Another motivation for upstream firms to openly disclose their knowledge is to improve their products by getting feedback and comments from downstream users. Downstream needs and feedback often give directions to industrial R&D. In this vein, the study of Cohen et al. (2002) on a sample of executives from US manufacturing firms showed that a majority of industrial R&D projects were initiated in response to information from buyers. Moreover, their results show that customers were also an important source of information for completion of R&D projects. Users are an important source of new knowledge since they tend to have better information about their idiosyncratic needs and the specific context of use of an innovation. In other words, users learn by using (Rosenberg, 1982). b. User Side In some cases, the end user of an innovation can also be an innovator. Frequently, downstream users develop, improve or adapt an innovation for in-house and/or custom use. This leads to the outcome of user-developed innovations (von Hippel 2002, Harhoff et al. 2003). Innovating users may be motivated to openly reveal information about an innovative product or process to induce manufacturer improvements. Using this information, suppliers may in turn improve upon the user-developed innovation, adapt their supplies to it or offer it at a price that is lower than the cost for users to produce it in-house. Manufacturers may have specific expertise to further improve the product and may be more able than users to render it more robust and reliable (von Hippel, 2002; Harhoff et al., 2003). If those improvements are made available for sale to all users on the market, the initial user-innovator may benefit from cheaper, more adapted and better quality inputs. In a game-theoretic model consisting of two user firms and one manufacturer, Harhoff et al. (2000) explored the conditions under which users who have developed an innovation would be willing to openly reveal the knowledge underlying their innovation. User firms have 1 In his model, Harhoff considers that the structure of the industry is exogenous when it is not determined by the sunk cost of R&D. 6

7 the choice between keeping their innovation secret and openly revealing it. The former case gives the firm a competitive advantage over the other firm, whereas the latter allows the manufacturer to learn about the innovation and to incorporate the improvements in the product it supplies to the user firms. They draw the conclusion that revealing the innovation may be an optimal strategy for the user firms. Therefore, firms may prefer to freely reveal their innovation to suppliers even if this leads to loss of comparative advantage since this loss may be offset by cost reductions and productivity increases triggered by new equipments. 2.2.Defensive Publication Firms may use knowledge disclosure through publication as a defensive strategy to prevent other firms from patenting an innovation (Parchomovsky, 2000; Johnson, 2004). Some authors have studied defensive publication in the patent race framework (Parchomovsky, 2000; De Fraja, 1993). A patent race, seen as a winner-take-all contest, assumes that when two firms are competing for a patent, the first to invent wins the race and the reward, namely the patent grant, while the loser must accept its defeat and leave the race with nothing. Within such a pattern, firms competing for a patent have all an interest in trying to win the race as conceding a patent to a competitor may lead to a substantial drop in profits for the loser or even leave the latter out of the market. In a theoretical model building upon the game theory, Parchomovsky (2000) attempted to illustrate the advantages of the pre-emptive publication strategy in the patent race framework. Stressing the possibility for firms to pre-emptively publish their research results, Parchomovsky reshaped the patent race model and argued that trying to win the race may not always be the profit-maximizing strategy. Rather, if a firm is lagging in the race, it may be optimal for the latter to prevent the leading firm from obtaining the patent by pre-emptively publishing its research results. This spoiler strategy is based on the fact that any publication becomes part of the prior art. Hence, since only inventions constituting a real inventive leap over the prior art are eligible for a patent, pre-emptive publication allows a firm to prevent a competitor from obtaining a patent by expanding the state of prior art which in turn affects the patentability of the rival s invention. However, the pre-emptive publication strategy may not always prevent a rival from patenting an innovation. Instead, it may even help the leading firm by accelerating its research effort and ease the issuance of the patent (Eisenberg, 2000). 7

8 Furthermore, De Fraja (1993) pointed out that if a firm may benefits from the technological success of a rival in a patent race, helping the rival to achieve this success by disclosing scientific knowledge might be a profit-maximising strategy for the firm. Besides preventing a rival from obtaining a patent, pre-emptive publication may also be useful to firms that do not wish to bear the cost of a patent application, but still want to keep the freedom to exploit their innovation. Johnson (2004) has compared defensive disclosure to patenting and secrecy and identified three cases where defensive disclosure is more likely to be optimal. First, defensive publications may be useful when a given innovation is not too technically challenging, that is to say, competitors may easily invent around the innovation. In this case, firms may want to avoid bearing the cost of a patent given that patents are likely to be inefficient as they could easily be circumvented. Also, secrecy would not prevent rivals to access to the innovation. Second, it can occur when a firm has to face rivals that leverage from patent litigation ( patent pirates ). Third, defensive publication is more likely to prevail when the patent office has not the ability to properly evaluate patent applications with respect to prior art and hence grants patents too easily. Finally, Johnson provides also evidence that defensive publication is usually practiced by larger firms since those firms are often have other competitive assets to rely on to secure profits (e.g. intellectual property portfolios, costs or scale advantages). 2.3.Collaboration, Networking and Knowledge Sourcing from Academia It is now widely acknowledged that the production of knowledge does not occur in isolation but is rather a collective process (Gibbons, 1994). Innovation has become a networkembedded process that results from several interactions involving many different actors (e.g. suppliers, buyers, universities, research institutes, banks, government agencies, etc.). In such context, it is crucial for innovative firms to collaborate and establish partnerships with other research organizations. Firms have an interest in collaborating with universities and other firms mainly for two reasons. First, collaboration increases the efficiency of the research by sharing the tasks, the costs and the risks of undertaking research (Muller and Pénin, 2006). The second and most important reason is that collaboration allows firms to gain access to technical opportunities and to external sources of knowledge (Hicks, 1995). Many studies have emphasized on the dependence of industrial innovation upon outside sources of knowledge and mostly upon the public science base (e.g. Mansfield, 1991; Adams, 2010) as 8

9 well as on the central importance for innovative companies to connect with the wider scientific community in order to stay abreast and to take advantage of the latest scientific developments (e.g. Hicks, 1995; Cockburn and Henderson, 1998; Zucker et. al., 1998). The world of scientific and technical knowledge exchange is governed by barter and research results can be seen as currency or bargaining chips in negotiations with potential collaborators (Hicks, 1995). In this respect, collaboration with academia is a central matter. In the case of contractual agreements with academic scientists, a conflict of interest regarding knowledge disclosure may rise between both parties given their respective institutional objectives and reward systems (Dasgupta and David, 1994). While firms tend to keep their knowledge secret to avoid the communication of valuable information to competitors, the academic system adheres to the norms of Open Science favouring the disclosure and the diffusion of research findings. When collaborating with academic partners, firms may impose restrictions regarding knowledge disclosure such as the delay of publications 2 (Blumenthal et al., 1997). However, in some cases firms may accept the disclosure of project outcomes as a concession in order to access to academic knowledge (Tijssen, 2004; Raffo and Simeth, 2011). Therefore, obtaining access to outside knowledge and resources may require concession and counterpart on behalf of the firm. And publication can be regarded as this concession. This reciprocal aspect of knowledge exchange relationships is also crucial in the case of informal interactions that are regarded as another important channel in knowledge exchanges with academia (Cohen et al. 2002). As said earlier, academic scientists share the norm of communalism that favours the disclosure of knowledge and its dissemination in the scientific community (Merton, 1973). This implies that researchers who do not contribute to the community cannot expect to take advantage of knowledge exchanges as academic scientists have little incentives to collaborate if the relationship is not reciprocal. Raffo and Simeth (2011) provided evidence that firms that rely on academic knowledge sources are more likely to publish and to do it more extensively regardless of the formality of their interaction with academia. Raffo and Simeth concluded that a need for academic knowledge sources leads to scientific contributions by firms, indicating that the access to these knowledge holders imposes the adoption of disclosure practices. Hence, firms wishing to leverage academic knowledge sources need thus to participate actively to the scientific 2 Firms may impose delay of publications for several purposes such as to allow for patent application, to protect their scientific lead, to slow the dissemination of undesired results, to allow time to negotiate a patent, or to resolve disputes over the ownership of intellectual property (Blumenthal et al. 1997). 9

10 community, and to participate they need to publish. Firms wishing to draw from external knowledge sources must also establish a reputation. This subject will be explored in the next subsection. 2.4.Publication, Credibility and Signalling In order to connect with potential collaboration partners and enhance knowledge exchanges, a firm has to be seen, that is, it has to build credibility and acquire a reputation (Hicks, 1995). Credibility and reputation are both established through competences. However, as it is the case in many markets, innovation occurs in an environment of imperfect and asymmetric information (Muller and Pénin, 2006). It might therefore prove difficult for firms to signal their technical and scientific competences to potential partners since those competences are not explicitly observable but rather closely related to tacit and unpublishable elements of knowledge 3. To the extent that information is incomplete and that competences are not directly observable, firms may want to generate signals in order to break the uncertainty with regards to their capabilities. In this respect, publications may be used as a signalling tool since they communicate useful scientific and technical information. In a survey on firm knowledge appropriability and acquisition, Levin et al. (1987) noted that publications were considered in many industries as an effective channel through which knowledge could be sourced. More specifically, publications can serve to signal competences to the outside world by signalling the possession of unpublishable and tacit resources (Hicks, 1995). Publishing helps thus to establish credibility. This credibility can be seen as the entrance ticket to knowledge exchange networks and as a necessary condition to link with external sources of knowledge. Moreover, papers can send signals that are useful in other markets such as capital, stock and labour market (Hicks, 1995; Muller and Pénin, 2006; Pénin, 2007). For instance, publication may allow companies to raise funds more easily on the capital market (Kinney et al., 2004; Pénin, 2007). Similarly, firms seeking to secure public grants, subsidies or government contracts may want to openly disclose their knowledge to show that they are able to use public funds effectively (Muller and Pénin 2006; Pénin, 2007). As we will see in the 3 A firm s knowledge base consists of codified knowledge, tacit knoweldge and knowledge that is embedded in equipments, intsruments, production plants, organisational structures and management routines. Codified knowledge refers to knowledge that has been reduced and converted to a written and transmittable form (e.g. patents, blueprints etc.). In contrast, tacit knowledge refers to knowledge that exists (sub)consciously in the human mind, is acquired through experience, imitation, and observation, and can be tranferred only by personal contact (Dasgupta and David, 1994, David and Foray, 1995). 10

11 next subsection, publication of scientific papers may also be used as an instrument to attract, recruit and retain highly skilled researchers and trained graduates. 2.5.Publication as mean to attract, retain and motivate researchers Firms can use scientific publications as a tool to recruit highly qualified researchers and the best-trained graduates and to motivate them (Stern, 2004; Hicks, 1995). Many scientists in industry often share the same norms and values than their academic counterparts. They want to be recognized by the scientific community for their discoveries and are concerned about their reputation (Sauermann and Stephan, 2010). Reputation and peerrecognition is regarded as crucial for a variety of reasons such as career advancements, receiving grants and receiving tenure. This leads researchers to value not only their wage but also the right to publish the results of their research since in the scientific community publication is the central instrument for obtaining recognition and receiving credit for the intellectual priority of scientific discoveries (Merton, 1973; Dasgupta and David, 1994). Publication rewards thus scientists via kudos. In this respect, firms may offer their researchers the concession to publish in order to sustain their motivation and to retain them. Therefore, publications may result from the firm s internal commitment to scientific practices driven by the pressure of its researchers to get the right to publish (Raffo and Simeth, 2011). Moreover, companies that encourage their researchers to publish are better positioned to attract the best scientists, the latter being usually reluctant to work in an environment where they cannot publish (Henderson and Cockburn, 1994). In the same vein, Scherer (1967) showed that the reputation of a lab, which is directly related to publication activity, affects the ability of the company to recruit researchers. Furthermore, the study of Stern (2004), based on a sample of PhD biologists, showed that researchers were ready to accept lower wages in return for the right to publish. Finally, as effort in research activities is difficult and costly to monitor (Cockburn et al., 2002), firms can use publication as a mean to measure researchers productivity and performance and to reward them cheaply. In this respect, the study of Henderson and Cockburn (1994) on a sample of pharmaceutical firms provided evidence that firms that were pro-publication, in the sense that they were using publications records or the researcher s reputation and standing in the scientific community as criterion for promotion, were more productive in drug discovery than their competitors. 11

12 2.6.Sector Conditions A firms decision to publish may be influenced by sector level characteristics. In some sectors, publications may be the result of requirements imposed by regulatory authorities 4. For instance, in areas such as pharmaceuticals or chemicals, firms are required to show scientific evidence in order to obtain the authorities approval to sell products on the market (Hicks, 1995). Moreover, the conditions of appropriability and the type of means regarded as effective to protect knowledge (e.g. patent, secrecy) in a sector may influence the firm s attitude with regard to publication. The effectiveness of the different means of appropriability tend to vary from one sector to another, hence the use of those means usually differ between sectors but is more homogeneous within an industry (Levin et al., 1987; Harabi, 1995). For instance, in the pharmaceuticals and chemicals industries, patents are seen as an effective instrument to protect product innovations 5. Hence, firms in those sectors tend to rely heavily on those protection means. It is important to distinguish between formal protection instruments, such as patents, and informal means, such as secrecy, as they reflect a different relationship with respect to knowledge disclosure. In a theoretical model, Gans et al. (2010) found that decisions to patent and publish are complementary. Among other results, they found that to the extent that the knowledge disclosed in patents and publications overlap, the incremental cost to the firm of disclosing through publications decreases once the knowledge has been disclosed through patents. Moreover, they showed that to the degree that a patent can effectively protect the firm from imitation, the negative consequences of additional disclosure through publication fall. In this respect, to the extent that patents are effective, it is assumed that firms that rely on such formal protection means will be more prone to publish. On the other hand, publication is less likely to occur in a sector where firms rely on secrecy to avoid any disclosure. Raffo and Simeth (2011) provided evidence that firms operating in sectors with an intensive use of 4 A good example is the Food and Drug Administration (FDA) in the U.S. : The FDA is responsible for protecting the public health by assuring the safety, efficacy, and security of human and veterinary drugs, biological products, medical devices, our nation s food supply, cosmetics, and products that emit radiation, and by regulating the manufacture, marketing, and distribution of tobacco products. (FDA website) 5 Levin et al. (1987) provide the following explanation to the effectiveness of patents in those sectors : comparatively clear standards can be applied to assess a chemical patent s validity and to defend against infringement. The uniqueness of a specific molecule is more easily demonstrated than the novelty of, for example, a new component of a complex electrical or mechanical system. Similarly, it is easy to determine whether an allegedly infringing molecule is physically identical to a patented molecule (Levin et al., 1987, p. 798). 12

13 formal protection instruments were more likely to publish whereas the opposite was found for those in sectors with a high use of informal protection mechanisms like secrecy. Furthermore, firms may be motivated to publish in order to develop an environment of openness in the sector where it operates and to encourage rival firms to do the same (Pénin, 2007). The study of Allen (1983) on the case of the British iron and steel industry in the nineteenth century provide a good example of openness between rival firms. Allen found that firms in this industry were regularly disclosing valuable technical information to competitors through informal disclosure and publication in the engineering literature. By sharing information about the design and performance of their plant, firms were able to build upon each other s work. This practice led to the achievement of several technical advances enabling productivity growth in this industry. The empirical study of Adams and Clemmons (2006) provided evidence that knowledge spillovers within an industry have a significant positive impact on the industry output of scientific papers. In contrast, Raffo and Simeth (2011) failed to find evidence that higher levels of intra-industry knowledge spillovers were encouraging or preventing a firm s attitude toward publication. Besides, they found that firms in sectors where a higher percentage of companies have published in the past were more likely to publish and to do it more extensively. In this literature review, focus has been put on the fact that firms may have many reasons to contribute to the building of publicly available knowledge by publishing articles in scientific journals. As we have seen, publications not only represent the firm s production of scientific and technical knowledge as a pubic good. It has been highlighted that publications, as instruments conveying knowledge, serve several purposes and may provide many indirect benefits to the publishing firm. Therefore, publications have to be placed in the context of the firm s whole corporate strategy and innovation process. It should be kept in mind that, given the strategic aspect of corporate research, firms manage carefully the release of their proprietary knowledge by balancing the costs and benefits of such strategy. Most of the explanations about the determinants of firms production and disclosure of scientific knowledge are mainly provided by the findings of theoretical studies. There has been to date little empirical evidence about the determinants of industrial scientific production. 13

14 3. Data and Sample Description The dataset used for the empirical analysis relies on two main data sources 6 referring to firms and their scientific publications. The first data source is the last edition of the EU Industrial R&D Investment Scoreboard released in October 2010 by the JRC-IPTS of the European Commission 7. The R&D Investment Scoreboard has been issued every year since 2004 and provides economic and financial data at the firm level for the top 1,000 R&D-active firms in the EU-27 and the top 1,000 outside the EU-27. Together these 2000 firms represent more than 80% of the total R&D performed in the private sector (BERD) worldwide (Cincera et al, 2010a). The information available in the R&D scoreboards is consolidated at the group level and includes, among others, R&D investments, net sales, number of employees, capital expenditures, the country where the company has its registered headquarters and the main business sector, based on the Industry Classification Benchmark (ICB) at the two digits level (i.e. 45 industry and services sectors). The second main data source is the Elsevier s SciVerse Scopus online database 8. Scopus is the largest abstract and citation database containing both peer-reviewed research literature and quality web sources. It offers access to over 18,500 active titles with more than 44 million records. The information concerning the publications 9 of each of the 1,400 top R&D firms has been retrieved manually 10 from this database using the Affiliation Identifier that allows to automatically identify and match an organization with all its research output. The research output of a firm may comprise various document type (i.e. article, article in press, conference paper, editorial, letter, review ), but for the sake of the analysis, only scientific articles have been considered as research outcome, as these tend to report original research results. Although the Scopus Affiliation Identifier is reliable, the dataset suffers from some limitations insofar as it does not cover the affiliates and other subsidiaries of the 1,400 parent companies. In fact it was not possible to retrieve manually the publications of all the firms affiliates as we don t have this information 11. Data on corporate research papers were consolidated at the parent company by including as much as possible corporate research laboratories, majority-owned subsidiary and other corporate affiliations. 6 See Appendix 1 for a summary of the data sources and variables See Appendix 2 for a sample record from Scopus. 10 Manual extraction represents a time consuming task as data export in Scopus is limited to a maximum of 2000 records per extraction 11 Cincera and Ravet (2011) did this exercise for all the EU companies listed in the 2008 edition of the R&D scoreboard. They found about 44,000 subsidiaries for a subset of 837 EU companies. 14

15 Our final data set consists of 779,563 papers published by 990 firms covering 45 industrial sectors 12 over a period spanning Descriptive Statistics Global trends in the output of corporate scientific research can be assessed from statistical analysis of papers published in peer-reviewed scientific and technical journals. The following analysis will try to illustrate, by means of a series of figures drawn from the data set we have built, the main trends and facts about scientific publication activity of the largest R&D companies worldwide. It is important to stress that the purpose of this bibliometric (i.e. literature-based) analysis is not to provide an in depth explanation of the evolution of corporate publishing in each country and/or sector but rather to provide an overall insight of the trends in the output of corporate scientific research Overall trends The first point to be made is that firms from the private industry publish extensively. This can be seen in Figure 1, which shows article production for the 990 sample firms over the period spanning During this period, those companies have published 768,995 papers, which represents on average a total of 11,831 papers per year. Overall, the results show that the number of papers published has significantly increased over time and that the long-term trend is one of growth. The important leap in the number of papers published between 1995 and 1996 is attributed to Sciverse Scopus treatment of pre-1996 content coverage 13. As pre-1996 data are less covered, the following analysis will put the focus on the period spanning See Appendix 3 for the list of the sectors covered by the data set. 13 For more details, see SciVerse Scopus Content Coverage Guide 15

16 40000 Figure 1 - Output of scientific papers ( ) Figure 2 focuses on the evolution of the production of scientific papers over the period. During this period, firms have published around 460,000 papers. Between 1999 and 2002, the results show a significant drop in corporate publishing that has decreased by 33%. We can attribute this significant fall to the systemic crisis that occurred at the dawn of the 21 st century. As a reminder, the world economies were hit by several crises, i.e. the bursting of the dot-com bubble followed by September 11 and the disarray of the financial markets. All the aforementioned reasons would precipitate major R&D-active companies to reduce the resources allocated to research and to engage in saving money programs in order to enhance their balance sheets. In other words, R&D budgets and investments were probably cut with, to some extent, an effect on publishing. From 2003 to 2005, the number of corporate research papers remained stable. Over the period, the number of paper rose by 40%. Finally, corporate publishing has dropped by 12% from 2008 to Overall, corporate publication output has decreased by 19% over the period. As in the case of the decrease that arises between 1999 and 2002, the drop that occurred between 2008 and 2010 may be interpreted as a consequence of the recent financial crisis that shook the world economy. In this respect, the OECD highlights that: US stock markets data suggest that companies have significantly reduced their R&D investments in the aftermath of the crisis (OECD, 2009, p. 24). The results reported by Cincera et al. (2010b) lead to a similar conclusion. 16

17 40000 Figure 2 - Output of scientific papers ( ) Industrial publications by countries Fig. 3 and 4 depict respectively the evolution of the number and the share of publication by country. For the sake of clarity, only the Top-9 publishing countries were represented on the charts. Together, these countries account for 95% of the papers of our dataset. The results show that the USA leads in terms of publication with no less than 177,210 papers published over the period spanning , which represents an average of 11,814 papers per year. US-owned firms represents between 35 and 40% of the papers published each year. Although the number of scientific papers published by US companies has dropped by 17% between 1996 and 2010, their share has remained more or less stable during this interval. Japan is the second largest country in terms of publication after the USA. Over the period, Japanese firms have published over 100,000 papers. The number of papers published by Japan-owned companies has dropped by 47% and their share has decreased from 27% in 1996 to 18% in The other countries to show a drop in publication output during this interval are France (-56%), the UK (-16%) and Germany (-7%). While the share of the UK and Germany have remained more or less constant (respectively 8% and 6%), the share of French companies has decreased significantly from 8,6% in 1996 to 4,7% in Among the countries that exhibit a growth in the number of papers, one can mention South Korea (+168%), Netherland (+129%), Denmark (+28%), and Switzerland (+22%). Switzerland overtakes France and Germany in the publication of scientific papers with a share that has 17

18 grown from 4,7% in 1996 to 7,1% in Although the share of South Korea, Netherland and Denmark have also increased over the same period, each of those countries still represent less than 5% of the scientific papers production Figure 3 - # of papers by country USA Japan France UK Germany Switzerland The Netherlands South Korea Denmark Figure 4 - % of papers by country USA Japan France UK Germany Switzerland The Netherlands South Korea Denmark 4.3. Industrial publication by sectors Figures 4 and 5 display the breakdown of corporate research papers by sector. Only the Top-10 publishing industries over the period spanning have been considered. Those sectors accounts for 70% of the papers published between 1996 and Among the 18

19 observed industries, the pharmaceuticals sector is by far the most prolific in terms of publications. Previous bibliometric studies have already shown that pharmaceutical companies publish extensively (Narin and Rozek, 1988; Koenig, 1983). Between 1996 and 2010, pharmaceutical firms have produced not less than 123,031 papers, which represents 27% of the papers published during this interval or an average of 8200 papers per year. The chemicals sector is the second largest producer of scientific papers with 35,732 papers published over the period. The chemicals industry represents on average 8% of the papers produced each year. Over the considered period, the output of scientific papers has dropped for almost all the observed sectors. Among the industries that show the highest decrease in the number of papers published, one can mention telecommunications equipement (-68%), fixed line telecommunications (-62%), computer hardware (-49%), general industrials (-44%), chemicals (-37%) and computer services (-31%). The decrease in the output of scientific papers was particularly sharp between 1999 and 2002, especially in sectors like telecommunications equipment, fixed line telecommunications, computer hardware and computer services. Those are mainly sectors related to information and communication technologies (ICTs) that were extremely exposed during the last Internet bubble crisis. For instance, the contribution fell from 6,6% to 2,6% for telecommunications equipment, from 6% to 3,8% for computer hardware, from 5,1% to 2,4% for fixed line telecommunications, The only sectors to experience a rise in the publication of scientific papers are semiconductors (+35%), electronic equipment (+15%) and biotechnology (+10%). 19

20 Figure 5 - # of papers by sector Pharmaceuticals (4577) Chemicals (135) Computer services (9533) Semiconductors (9576) Telecommunications equipment (9578) Computer hardware (9572) Biotechnology (4573) Fixed line telecommunications (653) Electronic equipment (2737) General industrials (272) Figure 6 - % of papers by sector Pharmaceuticals (4577) Chemicals (135) Computer services (9533) Semiconductors (9576) Telecommunications equipment (9578) Computer hardware (9572) Biotechnology (4573) Fixed line telecommunications (653) Electronic equipment (2737) General industrials (272) 20

21 4.4. Quality of Knowledge Another question that arises is whether scientific papers from the private industry contain valuable information. In this respect, citation counts indicate the influence and relevance of prior research and hence provide a measure of the quality of the knowledge disclosed through scientific papers. Therefore, the observation of citation patterns allows us to assess to some extent whether corporate scientific publications provide useful knowledge. Besides reflecting the relevance of scientific research performed by companies in a country or an industry, a high number of citations could mean that papers are regarded as an important channel through which knowledge is sourced. In this vein, citations are also used as an indicator of knowledge flows. Unfortunately, the data on scientific papers extracted from Scopus only provides the number of citations received by each paper but does not allow us to observe the origin of the citation which could provide useful information regarding the various knowledge flows (self-citation vs. non-self-citation, academic/public vs. corporate/private, inter-industry vs. intra-industry, same scientific field/discipline vs. different field/discipline, patent vs. publication). Figures 7 and 8 depict respectively the average number of citations per paper by country and by sector. Not surprisingly, scientific papers published in earlier years usually have more citations per paper than papers published in recent years 14. As the results show, papers published in the biotechnology sector are the most cited with an average of 47,5 citations per paper per year over the period spanning Biotechnology is then followed by pharmaceuticals (29,6), computer services (23,1), telecommunications equipment (19,6), chemicals (16,8), computer hardware (13,7), fixed line telecommunications (12,7), electronic equipment (12,4), semiconductors (12), and general industrials (11). Papers published by Swiss companies are the most cited with a mean of 30,7 citations per paper per year over the period. Switzerland is followed by the USA (24,9), Denmark (24,8), UK (23,6), France (19,3), Germany (15,1), Netherlands (13,4), Japan (11,2), and South Korea (8,9). 14 This is because earlier papers have a greater opportunity to be cited than recent ones. For instance, papers published in 2001 have a 10 years citation period (from 2001 to 2010 included) while papers published in 1996 have a 15 years citation period (from 1996 to 2010 included). Consequently, the number of citations drops as the length of the citation period decreases. As we computed the average number of citations per paper for each year, this measure is smaller for recent papers because the number of citations is low compared to the number of papers published. 21

22 Figure 7 - Average number of citations (per paper) by sector Pharmaceuticals (4577) Chemicals (135) Computer services (9533) Semiconductors (9576) Telecommunications equipment (9578) Computer hardware (9572) Biotechnology (4573) Fixed line telecommunications (653) Electronic equipment (2737) General industrials (272) 60 Figure 8 - Average number of citations (per paper) by country USA Japan France UK Germany Switzerland The Netherlands South Korea Denmark 00 22

23 4.5. Collaboration Trends in co-authorship patterns of research publications are also of interest as joint authorship often reflects collaborative research activity as well as generation and exchange of new or current knowledge (Henderson and Cockburn, 1998). In this respect, co-authorship data may be used as an indicator of collaboration in corporate scientific research. The role of collaboration in the production of scientific knowledge has substantially grown over time. According to the OECD, the trend in scientific knowledge production is shifting from individual to group, from single to multiple institutions, and from national to international (OECD, 2009, p. 114). OECD figures show that both domestic and international coauthorships have been increasing rapidly over the past decade with domestic co-authorship becoming the most important form of scientific collaboration 15 (OECD, 2009). Figure 9 depicts the average number of co-authors per paper by sector. As previously, only the Top-10 publishing industries over the period spanning have been considered. Overall, the results exhibit an increase in the average number of co-authors over the period spanning It should be mentioned that this trend is observed for all the sectors included in our data set where the mean number of authors per papers has increased from 4,4 to 6,6 over the interval The average number of co-authors per paper is the greatest for the biotechnology and pharmaceuticals industries. Those two sectors also exhibit the largest rate of growth in the average number of authors per paper over the period. In the biotechnology industry the average number of co-authors per paper has increased from 6 in 1996 to 9,9 in 2010, whereas in the pharmaceuticals sector this number has increased from 5,5 in 1996 to 8,8 in This growth of co-authorship may be partly attributed to the development of the information and communications technologies and to the widespread use of the Internet, which reduced the opportunity cost of time in collaboration. Collaboration in the production of scientific knowledge is thus growing in importance over time, especially in sectors where research is interdisciplinary. For instance, biotechnology is an industry where research occurs at the interface of several disciplines and involves many distinct expertises and competences. Hence, various skills and knowledge are required to work on research projects in this area. 15 Domestic co-authorship refers to collaboration between researchers of different institutions in the same country, while international co-authorship correspond to collaboration between researchers from different countries (OECD, 2009). 23