Mapping private R&D outputs: The contribution of top R&D companies to scientific literature 1

Transcription

1 Mapping private R&D outputs: The contribution of top R&D companies to scientific literature 1 Roberto Camerani a, Daniele Rotolo a, Nicola Grassano b a SPRU, University of Sussex, Brighton, UK b JRC-IPTS, Seville, Spain Paper submitted to: SPRU s 50th anniversary Conference Brighton, 7-9 September, Preliminary paper, please do not cite without author s permission 1 We wish to express our gratitude to JRC-IPTS (Joint Research Centre - Institute for Prospective Technological Studies) for commissioning a study, the data of which is used in this article. The views expressed here are our own and do not reflect that of the JRC-IPTS. We also wish to acknowledge the contribution of Jack Low and Richard Duffy who helped extensively in the data collection. 1

2 INTRODUCTION The present paper examines the extent to which top R&D companies are involved in publishing scientific papers. Publication data are a source of bibliographic information that can be exploited to build indicators capable of generating additional perspectives on or more precise knowledge of firms R&D output and innovation dynamics as compared to most conventional R&D data such as financial and patent data. Publication data can also inform about the knowledge areas in which firms R&D efforts are focused, on the impact of firms R&D on subsequent scientific and technological developments, on R&D collaboration activity, and on the geographical location and distribution of firm s R&D activities. These types of evidence can provide policymakers with strategic intelligence for the development and assessment of R&D policy instruments (e.g. Debackere & Glänzel, 2004; Martin, 2016; Rotolo, Rafols, Hopkins, & Leydesdorff, 2016). To our knowledge, a limited number of studies have examined firms publication activity. These have provide evidence that, in certain sectors, firms publish a significant amount of their R&D outcomes (Li, Youtie, & Shapira, 2015; Rafols et al., 2014). For example, pharmaceutical and electronic companies were found to publish scientific articles as much as medium sized universities (Hicks, 1995). Youtie and Kay (2014) found that about 8% of publications in the emerging field of nanotechnology have a corporate author. Other studies have investigated the impact of scientific publication on firm performance. For instance, Simeth and Cincera (2015) found a positive relationship between firms scientific activities (including publishing) and their market value, in the context of high-tech sectors. But, why do firms publish? It is true that publishing may limit the ability of firms to appropriate the value of their R&D efforts through other intellectual property protection mechanisms (e.g. patents), but it is also true that the contribution to scientific literature enables firms to be involved with the academic community. This, in turn, can provide firms with learning opportunities and access to knowledge and resources as well as enhances firms reputation and legitimates firms research work (Ding, 2011; Jong & Slavova, 2014). Previous research has been mostly limited to specific sectors, such as science-intensive, hightech, or manufacturing, or have focused on the simple counting of publications by year, thus overlooking the rich set of information that could be extracted from publication bibliographic data. This paper contributes to our understanding of firms publication activity by examining this phenomenon in a systematic manner across industrial sectors and by using the bibliographic information of publication data of these companies to generate more granular information about their publication strategies. METHODS AND DATA We rely on a large sample of firms intensively engaged in R&D. The sample is composed of the top 2,500 worldwide companies in terms of R&D investments, as listed in the 2014 EU Industrial R&D Investment Scoreboard 2 (see Table 1). These companies are classified in

3 sectors (on the basis of Industry Classification Benchmark ) and account for about 90% of the global private R&D investments. To ensure coverage in the collection of publication data, our sample also accounts for 569,919 subsidiaries of these companies. In order to identify publications authored by firms (including their subsidiaries) in our sample, we searched for firms names in authors affiliation addresses listed in publication data. To do so, we developed an approach capable of maximizing the data collection coverage (i.e. recall) while reducing the number of false positive cases (i.e. increasing precision). Table 1. Top 2,500 worldwide companies in terms of R&D investments by sector. (Industry Classification Benchmark) Number of firms Total number of subsidiaries Aerospace & Defence 51 11,699 Alternative Energy Automobiles & Parts ,384 Banks 29 42,924 Beverages 9 5,686 Chemicals ,130 Construction & Materials 72 43,911 Electricity 29 22,032 Electronic & Electrical Equipment ,663 Equity Investment Instruments 1 22 Financial Services 15 2,286 Fixed Line Telecommunications 20 19,000 Food & Drug Retailers 4 5,263 Food Producers 60 14,945 Forestry & Paper 13 3,233 Gas, Water & Multi-utilities 13 17,787 General Industrials 95 34,402 General Retailers 18 16,801 Health Care Equipment & Services 97 15,407 Household Goods & Home Construction 38 9,804 Industrial Engineering ,939 Industrial Metals & Mining 41 10,074 Industrial Transportation 11 11,893 Leisure Goods 38 6,635 Life Insurance 5 2,118 Media 18 3,060 Mining 11 2,676 Mobile Telecommunications 9 4,592 Nonequity Investment Instruments 1 1,069 Nonlife Insurance Oil & Gas Producers 27 28,000 Oil Equipment, Services & Distribution 16 3,243 Personal Goods 47 7,927 Pharmaceuticals & Biotechnology ,691 Real Estate Investment & Services 3 30 Software & Computer Services ,197 Support Services 32 5,549 Technology Hardware & Equipment ,553 Tobacco 6 2,018 Travel & Leisure 24 14,251 Total 2, ,919 Source: Authors elaboration on the basis of the 2014 EU Industrial R&D Investment Scoreboard data. 3

4 Firm names cleaning We first cleaned the names of the firms in the sample. This activity involved the following three steps. First, we identified name variations or acronyms of the 2,500 firms included in our sample by a desktop search. For example, for the case of the firm SVENSKA CELLULOSA AB SCA we found three name variations: SVENSKA CELLULOSA AKTIEBOLAGET, SWEDISH CELLULOSE COMPANY, and SCA. We also made sure that companies mainly known by their acronym had their full name listed (e.g. IBM and International Bus iness Machines ). Overall, we identified 1,384 name variations for the 2,500 firms included in our sample and specifically at least one name variation for 1,084 (about 43%) of the firms in our sample was added. Second, we used the group names and their variations to thin out the list of subsidiaries. In particular, we removed all the subsidiaries the name of which included one of the names of their parent firms. Publication co-authored by researchers in these subsidiaries will be retrieved by searching for parent firms names in authors affiliation addresses (e.g., KELLOGG S PRODUITS ALIMENTAIRES was not included since the related query already searches for the parent firm s name, i.e. KELLOGG ). Similarly, national branches having the same name of the parent company have been removed (e.g. Philips France, Philips Italia, etc.) Finally, we cleaned remaining subsidiaries names by combining an automatic cleaning procedure with manual screening. Specifically, we developed a routine based on regular expressions to remove those portions of the text of subsidiary names that are likely to reduce the recall when retrieving publication data. These include, for instance, symbols and punctuation (e.g.,, :, ;, # ), and double spaces. Specific attention has been dedicated to business entities and abbreviations (e.g., Ltd, Inc, GmbH, AG ). In some cases, these business entities do not appear in the addresses of the publications (e.g., the address includes simply Apple, instead of Apple Inc. ). Therefore, we identified and compiled a list of 159 business entity extensions and we removed them from the subsidiary names. As a result, this process returned a list of about 350,000 subsidiaries on a total of almost 570,000. In order to ensure that the automatic cleaning was not too aggressive (e.g. SCA is both a firm s name - Svenska Cellulosa Aktiebolaget - and a French business entity abbreviation - Société en commandite par actions ), we manually checked the remaining 350,000 subsidiaries. In doing so, we performed additional cleaning: for example, given that the geographical location of each subsidiary was available and that we can use this information to build a more precise query, we removed (when possible) country names from the names of the subsidiaries (e.g., GOLLEK ARGENTINA S.R.L was revised as GOLLEK ). Subsidiaries with common names were flagged (e.g. plant or computer systems ) and used to increase the precision of the queries or to support the subsequent cleaning of publication raw data as discussed below. Building the queries We used the cleaned group and subsidiary names to query Web of Science (WoS) by Thomson Reuters. The general structure of the queries used to retrieve firms publication data is reported below: 4

5 (AD=( Firm s name ) OR AD=( Firm s name variation #1 ) OR AD=( Firm s name variation #2 ) OR OR AD=( Subsidiary s name #1 SAME Subsidiary s country ) OR AD=( Subsidiary s name #2 SAME Subsidiary s country ) OR ) AND (PY=(2011 OR 2012 OR 2013 OR 2014 OR 2015)) We searched for the names in the full address field ( AD ). We also tested other possibilities, such as the using the organization field ( OO ), or the Organization-Enhanced field ( OG ), but they did not prove to be reliable enough for the purposes of our study. Since all the groups in the sample are internationally known, we searched for group names (and their variations), without imposing any country limitation. In the case of subsidiaries, we searched for their names only in the country in which they were located using the Boolean code SAME in the advanced search interface of WoS. Finally, we restricted the search to the years between 2011 and 2015, and we produced one query for each group in our sample. Publication data collection We then relied on the WoS Core Collection to retrieve publication data. The general strategy has been to run the query and download all the publications that WoS returned for each group, including full bibliographic detail and citations. However, when WoS returned an unexpectedly high number of publications (we used a threshold of 5,000 publications) for a given firm, we manually examined the query and the retrieved publications further. We specifically checked the presence of subsidiaries with common names (e.g., Life ) in the query and the extent to which these were returning a large number of false positive records, i.e. records not authored by researchers at the firms and subsidiaries included in the sample. Query were revised for more precise searchers when a high number of false positives was identified. Publication data cleaning Our data collection approach tends to maximise recall of publications to the detrimental of precision. Therefore, the raw data are still likely to include a high number of false positives. For this reason, the raw publication data went through a process of cleaning by means of an automatic procedure that examined the text reported in the C1 field of WoS data. First, we excluded all publication records the affiliation addresses of which included only universities, hospitals (indicated as Univ and Hosp in WoS affiliation address data, respectively), and other organisations not included in our sample of firms and subsidiaries: 3 3 We identified 80 cases of organizations that could be excluded with this approach. 5

6 Case 1 Univ St Etienne, Corneal Graft Biol Engn & Imaging Lab, EA2521, SFR 143, Fac Med, St Etienne, France; Univ Hosp St Etienne, Dept Pathol, St Etienne, France; Inst Univ France, Paris, France Second, we excluded all publication records that were retrieved by the query because subsidiaries had the same name of the cities listed in the authors affiliation addresses (some publications that were erroneously selected because of some subsidiaries names have city names): Case 2 Subsidiary s name: HALIFAX Affiliation addresses: Univ Bern, Univ Hosp Bern, Dept Anaesthesiol & Pain Med, Bern, Switzerland; Calderdale Royal Hosp, Emergency Dept, Halifax HX3 0PW, Salterhebble, England; This enabled us to remove 526,593 false positive records of publications (about 41% of the initial dataset). The final dataset of publications included 747,288 publications which have been the base for the analyses carried out in this paper. The next step in the cleaning process of the data will involve a manual screening of the publications. PRELIMINARY ANALYSES AND RESULTS The publication data collected were processed by mean of a routine to produce the set of indicators describing the publication activity of firms included in our sample. This section will show some of these indicators and some preliminary results. The first remarkable result is that all the 40 sectors included in our sample feature some publication activity. However, for simplicity, the following tables will refer to the publications of the top 10 sectors in terms of each of the indicator considered. Table 2 summarizes the number of scientific publications produced in the top 10 sectors (in terms of number of publications), the average number of publications per firm, and the proportion of publishing firms. While the top publishing sectors comprise the usual suspects (e.g. Pharmaceuticals & Biotechnology, and Chemicals), these also include some less expected sectors such as Electronic & Electrical Equipment and Software and Computer Services. The publication activity does not seem to be a prerogative of a few firms, but quite common among firms actively engaged in R&D. At least 80.6%, of the firms in the top 10 sectors, in fact, published at least one publication in the observation period. The average number of publication per firms varies greatly across sector, ranging from around 100, to more than

7 Table 2. Summary statistics of firms publication activity of the top 10 sectors in terms of number of publications ( period). Number of firms Number of publications ( ) Average number of publications per firm Proportion of publishing firms 1 Pharmaceuticals & Biotechnology , % 2 Electronic & Electrical Equipment , % 3 Software and Computer Services , % 4 Technology Hardware & Equipment , % 5 Automobiles & Parts , % 6 Chemicals , % 7 General Industrials 95 42, % 8 Health Care Equipment & Services 97 34, % 9 Industrial Engineering , % 10 Construction & Materials 72 18, % Table 3 compares the rank position of a sector with respect to the overall number of publications firms in the sectors produced from 2011 to 2015 with the average growth rate of publication activity in the same observation period. The table reports the top 10 sectors in terms of publication growth rate. In this regard, the sectors characterized by the most prominent growth in the publication activity are Construction & Materials (about 45%), Travel & Leisure (about 26%), Electronic & Electrical Equipment (about 15%), Mining (about 11%) and Mobile Telecommunications (about 11%). In general, if we exclude Electronic & Electrical Equipment, none of the top publishing sectors is also fast growing in terms of publications. This may indicate the existence of a sort of catching up process of relatively less publication intensive sectors with respect to the most intensive ones. Table 3. Top 10 sectors in terms of average publication growth rates ( period). Publication activity Publication rank position growth rate 1 Construction & Materials Travel & Leisure Electronic & Electrical Equipment Mining Mobile Telecommunications Oil Equipment, Services & Distribution Financial Services Beverages Electricity Forestry & Paper Table 4 compares the rank position of a sector with respect to the overall number of publications firms in the sector produced from 2011 to 2015 with the average number of authors involved in publications authored by firms of a sector. The first consideration regarding these results is that collaborations and co-authoring seems to be extremely common, with the average number of authors ranging from 12 to 96. These co-authorships, however, vary significantly within sectors as evidenced by high standard deviations. The 7

8 sectors characterized by the largest R&D teams are Electronic & Electrical Equipment (on average about 96 authors per publication), Technology Hardware & Equipment (on average about 64 authors per publication), Chemicals (on average about 54 authors per publication), Equity Investment Instruments (on average about 53 authors per publication), and Industrial Metals & Mining (on average about 41 authors per publication). The results highlighted in the next two tables will further elaborate the phenomenon of collaborations and co-authorship by showing how many organizations from different countries are involved in these collaborations, and the proportion of publications in which at least one university is involved. Table 4. Top 10 sectors in terms of average number of authors involved in publications authored by firms in each sector (including standard deviation and maximum). Publication activity rank position Number of authors Average Std. Dev. Max 1 Electronic & Electrical Equipment ,900 2 Technology Hardware & Equipment ,154 3 Chemicals ,154 4 Equity Investment Instruments ,060 5 Industrial Metals & Mining ,154 6 Electricity ,062 7 Software and Computer Services ,154 8 Alternative Energy ,154 9 Beverages Oil & Gas Producers ,828 Table 5 compares the rank position of a sector with the distinct number of countries of the affiliations involved in the publications firms in the sector produced from 2011 to In general, we can observe an intense international collaboration activity, with up to 208 distinct countries involved. The sectors characterized by the broadest variety of countries are Alternative Energy, Industrial Engineering, Nonequity Investment Instruments, Technology Hardware & Equipment, and General Industrials. Five out of the top 10 sectors for geographical variety are also in the top 10 for number of publications. Table 5. Top 10 sectors in terms of number of distinct countries (geographical variety) involved in publications authored by firms in each sector. Publication activity Geographical variety rank position (number of countries) 1 Alternative Energy Industrial Engineering Nonequity Investment Instruments Technology Hardware & Equipment General Industrials Food & Drug Retailers Chemicals Software and Computer Services Industrial Metals & Mining General Retailers

9 Table 6 compares the rank position of a sector with the proportion of publications of the sector that involve at least one university from 2011 to Overall, collaboration with academia is extremely frequent. All sectors in the top 10 have more than 70% of their publications involving one or more universities (and even considering sectors not included in the top 10, this percentage is quite high, and above 47%). For some sectors, such as Beverages, Industrial Transportation, Personal Goods, and Oil & Gas Producers, this proportion is even higher and above 75%. With the exception of Technology Hardware & Equipment, none of the top sectors for university collaborations are also in the top 10 for publication activity. Table 6. Top 10 sectors in terms of proportion of publications authored by firms in each sector ( period) that involved at least one university. Publication activity Proportion of publications rank position involving universities 1 Beverages Industrial Transportation Personal Goods Oil & Gas Producers Travel & Leisure Technology Hardware & Equipment Equity Investment Instruments Life Insurance Mining Leisure Goods Finally, table 7 compares the rank position of a sector with the proportion of publications of the sector that acknowledged external funding from 2011 to The importance of external sources of funding seems to differ significantly across sectors, ranging from about 48% to 61%. However, certain sectors, such as Life Insurance, Equity Investment Instruments, Personal Goods, Industrial Transportation, Mining, and Fixed Line Telecommunications, seem to rely on more external financial support than others, with more than 50% of their publications acknowledging support from at least one funder. Quite interestingly, none of the top publishing sectors are in the top ranking for acknowledgement of external funding. Table 7. Top 10 sectors in terms of proportion of publications authored by firms in each sector ( period) that reported acknowledgements to external funding. Publication activity Proportion of publications rank position acknowledging funding 1 Life Insurance Equity Investment Instruments Personal Goods Industrial Transportation Mining Fixed Line Telecommunications Oil & Gas Producers Beverages Electricity Leisure Goods

10 CONCLUSIONS This paper aimed at mapping private R&D outputs by showing the publication activity of top R&D companies. In addition, by capitalizing on the richness of the rich bibliographic information included in publication data, the paper has also developed indicators that provide additional insights about the R&D efforts and strategies of the most active R&D companies. We started with the list of the 2,500 included in the 2014 Scoreboard (also including about 570,000 subsidiaries of these firms), and developed a novel methodology to retrieve, collect, and clean firm s publication data. We focused on the period. The main results of this endeavour was a dataset with more than 747,000 publications authored by firms included in the sample for which we also collected full bibliographic details (e.g., authorship, citations, scientific areas). We used these data to calculate the number of publications authored by each firms and to produce several R&D indicators. These indicators regarded different aspects of firms publishing activity, such as the growth of the number of publications over the observation period, co-authorship, academic collaborations, geographical variety of the collaborations, and external funding. These indicators were used to examine firms publication activity at sectoral level. The following main findings can be highlighted. First, publish scientific articles in academic journals seems to be a very common activity across firms in the sample. About 75% of the firms authored at least one publications, and in many cases they contributed to hundreds or even thousands publications in the period. Second, corporate publishing is not only a prerogative of science-based sectors such as Pharmaceutical, Biotechnology and Chemistry, but it is widespread across sectors. Third, as in the case of academic research, collaboration seems to have a fundamental role also in firms publication activity. In particular, we found that co-authorship is a widespread phenomenon (publications are authored by multiple researchers in most of the cases), that firms often engage in international collaborations involving organisations based in a broad range of countries, and finally, that the collaborations between the firms in our sample and universities is relatively intense (at least 47% of publications authored by researchers employed by the firms in our sample involved at least one university). Fourth, the ability to attract external funding does not seem to be linked to the ability to publish per se. None of the top publishing sectors, in fact, are also in the top ranking for acknowledgement external funding In conclusion, we believe this study provides two original contributions. The first one is to the literature on corporate publishing. Extant research has mainly focused on specific sectors such as Pharmaceutical and Biotechnology, thus paying limited attention to other sectors. Our study provided a systematic overview of the extent to which firms across sectors publish. The second contribution is methodological. Methods and routines developed in order to clean firms/groups and subsidiaries names, to match these names with authors addresses in WoS data (and potentially with other data sources), to build queries for retrieving publication data with high precision and recall, and to clean the collected data, represent a considerable methodological contribution. This could be potentially reused and applied to other contexts and areas. 10

11 REFERENCES Debackere, K., & Glänzel, W. (2004). Using a bibliometric approach to support research policy making: The case of the flemish BOF-key. Scientometrics, 59(2), doi: /b:scie Ding, W. W. (2011). The Impact of Founders Professional-Education Background on the Adoption of Open Science by For-Profit Biotechnology Firms. Management Science, 57(2), Retrieved from Hicks, D. (1995). Published papers, tacit competencies and corporate management of the public/private character of knowledge. Industrial and Corporate Change, 4(2), doi: /icc/ Jong, S., & Slavova, K. (2014). When publications lead to products: The open science conundrum in new product development. Research Policy, 43(4), doi: /j.respol Li, Y., Youtie, J., & Shapira, P. (2015). Why do technology firms publish scientific papers? The strategic use of science by small and midsize enterprises in nanotechnology. The Journal of Technology Transfer. doi: /s Martin, B. R. (2016). R&D policy instruments a critical review of what we do and don t know. Industry and Innovation, doi: / Rafols, I., Hopkins, M. M., Hoekman, J., Siepel, J., O Hare, A., Perianes-Rodríguez, A., & Nightingale, P. (2014). Big Pharma, little science?: A bibliometric perspective on Big Pharma s R&D decline. Technological Forecasting & Social Change, 81, doi: /j.techfore Rotolo, D., Rafols, I., Hopkins, M. M., & Leydesdorff, L. (2016). Strategic intelligence on emerging technologies: Scientometric overlay mapping. Journal of the Association for Information Science and Technology, in press. doi: /asi Simeth, M., & Cincera, M. (2015). Corporate Science, Innovation, and Firm Value. Management Science, Published online in Articles in Advance 20 Oct Youtie, J., & Kay, L. (2014). Acquiring nanotechnology capabilities: role of mergers and acquisitions. Technology Analysis & Strategic Management, 26(February 2015), doi: /