To some, scientific research produces discoveries and information which can be codified into documents that circulate easily around the world. Such information then helps guide applied research. Others emphasize that research depends upon and generates an interwoven stream of heterogeneous assemblages comprising codified knowledge in papers, but also embodied skills and craft knowledge, know-how, laboratory techniques, biological substances, materials, software and equipment. The resources are varied in nature and interlinked in complex ways making them hard to separate. They also emerge and evolve continuously. Except for information contained in papers, these resources are not freely available (Hilgartner, 1994). When the localized resources do move, traditionally they have been exchanged in relationships that fall outside markets or organizations (Collins, 1982). These relationships constitute a network form of organization. Callon argues that the most important result of research is to produce new heterogeneous networks, such as those formed by the relationships between researchers exchanging their resources (Callon, 1995).
This suggests that to understand the science and technology relationship, we need to know more about the networks through which embodied skills and tacit knowledge, materials and substances are exchanged. The networks organizing knowledge exchange facilitate innovation and fall outside markets or firms. Freeman argues that external sources of information accessed by firm personnel through networking have long been found to be important for successful innovation. There is a paucity of evidence because networks are mostly informal and so extremely difficult to trace and to analyze (Freeman, 1991). Although informal, the OECD considers them extremely important for diffusing the non-codified components of knowledge (OECD, 1996).
Even when networking relationships are strengthened into research partnerships, a great many remain informal, with no systematic way to track these partnerships or study them in detail (Hagedoorn et al., 2000). For those that reach the formalized stage, strategic alliance databases such as those maintained by Hagedoorn, Link and Vonortas have been developed into an excellent tracking tool. We argue here that bibliometric indicators, particularly in combination, may provide another method of tracking networks of innovation, one which gets closer than the strategic alliance databases to the informal networks and partnerships that are otherwise so difficult to trace.
It seems logical to surmise that formal R&D alliances if they are to produce anything of substance, should produce some jointly owned patents. Therefore we began our search for alliance-bibliometric relationships with co-assigned patents. We have nothing to build upon because co-assigned patents apparently have not been studied previously. Here, we can only make a small contribution to filling this vacuum. The singular characteristic of coassigned patents is that they are so rare, which probably accounts for our ignoring them. Companies seem to have a positive aversion to sharing their intellectual property, although some may be moving beyond this as Figure 1 illustrates. Figure 1 displays the share of US invented US patents coassigned. For this count, individual inventor patents were removed and parent-subsidiary joint patenting was removed for the largest patenting companies. Domestic coassigned patents accounted for about 0.2% of US invented patents owned by companies and universities in the early 1980s. Beginning in 1984 this percentage began to climb and reached 1.4% by 1998/99. In absolute terms the number of coassigned patents increased from about 50 in the early 1980's to 875 in 1998/99.
It is hard to believe that so few patents could be economically significant, but we should be cautious in drawing this conclusion. Since the value of patents is so unevenly distributed, with many worthless patents and a few of high value, if joint patents were of extremely high value, they might well carry an economic weight belied by their numbers. Logically, very high value inventions might be more likely to be jointly patented as the high value of the invention might help companies overcome their natural aversion to joint intellectual property ownership. A check on the citation characteristics of the patents suggests that joint patents are in fact not more valuable than other patents. CHI defines the CII or "current impact index" as the number of times the previous five years of patents are cited in the current year, relative to all patents in the U.S. patent system. A value of 1.0 represents average citation frequency; a value of 2.0 represents twice average citation frequency; and 0.25 represents 25% of average citation frequency. The Current Impact Index for the coassigned patents is 1.16 compared with 1.19 for all US invented patents assigned to companies or universities suggesting that coassigned patents are no different in impact from patents in general.
The rate of coassignment varies across technologies. In chemicals, for example, almost no patents are coassigned (0.08%), while in biotechnology about 7% are coassigned. The growth in coassignment is also uneven. In some areas, coassignment has taken off; in others, the rate has not changed in 20 years. In 1980, excluding biotechnology which always had a higher rate of coassignment, the maximum share of patents coassigned in any one of CHI's 30 technology classification was 1.43% and the minimum was 0. By 1999, the maximum had risen to 7.1% while the minimum remained at 0%.
Figure 2 illustrates these differences between technologies, plotting the share of patents coassigned in 1999 against the right-hand axis using the line and the public-private composition of coassigned patents against the left hand axis using the bars. Technologies with a high rate of coassignment on the figure are also those in which coassignment has grown since 1980. In health technologies, the public sector (universities and government laboratories) are involved in a high percentage of coassigned patents and the growth in coassignment may be driven by the growth in university patenting.
The figure suggests areas for further investigation. What accounts for the huge difference in coassignment between pharmaceuticals and chemicals? What has happened in glass, clay and cement and in aerospace to make co-patenting so attractive? The figure also suggests that biotechnology, whose coassignment rate was always high, is unusual only in that public sector institutions patent heavily in this area, and as they do with their papers, they are much more willing to share ownership of their patents than are companies.
In addressing the question of the relationship between strategic alliances and bibliometric data, we broadened the remit beyond joint patents, indeed beyond patent data to include eight of the ten possible paper and patent linkage dimensions. We pursued this question through a case study of a successful biotechnology firmChiron. This case study was undertaken as part of a collaboration with Woody Powell of Stanford University who has supplied information on strategic alliances.
Constructing this analysis was an intricate and time consuming task, as three databases had to be cleaned and aligned: alliances, the CHI's Science Literature Indicators database of papers and citations constructed for NSF using the Science Citation Index and CHI's patent indicators database. Eight bibliometric dimensions of citation and collaborative production were examined. We will use the term "linkage" to refer collectively to the set of relationships we are working with namely: co-authorship, co-patenting, referencing and citation. The analysis amounts to examining the science and technology network from the perspective of Chiron. In essence we ask which institutions produced the papers and patents referenced by or citing to Chiron papers and patents, and which institutions co-authored papers or co-patented with Chiron. Details of methodology are described in Appendix A.
The eight dimensions of science & technology linkage information are easy to mix up, so in an effort to keep things straight, the notation in table 1 will be used to label them.
|pub||SCI indexed papers, publications|
|citation relationship, arrow points from the referencing to the cited document|
The eight dimensions are:
Chiron publication links
Chiron patent links
The result of the analysis is a table of institutions linked to Chiron in one or more dimensionssee Figure 3. The linkage data is reported as ranks in each dimension; these are obtained by sorting the institutions descending by number of links in the dimension and then assigning ranks. The list is ordered first descending by number of dimensions in which an institution is linked to Chiron and then ascending by the sum of the ranks across dimensions. In addition to linkage information, the table lists the number of life science papers from the institution and the number of patents to facilitate a normalized perspective on the strength of linkage. For example, Creative Biomolecules is very highly ranked, and the publication and patent data reveal how small the organization is compared to other highly ranked organizations.
Is there a correlation between formal R&D alliances and the bibliometric dimensions? Logically, we might expect to find a relationship, particularly between R&D alliances and joint patenting. The joint development of intellectual property that presumably occurs in an R&D alliance should result in joint ownership. However, research forthcoming from Hagedoorn demonstrates the opposite, namely that joint patents and R&D alliances are uncorrelated. In our case study we ran a similar exploratory analysis on Chiron's data. Taking organizations that have at least one paper link, one patent link and one R&D alliance, and controlling for publishing and patenting size, we find no correlation between R&D alliances and any bibliometric dimension. This means that none of the bibliometric dimensions can be used to predict the number of alliances Chiron has with an organization.
However, there is another question that can be asked, namely: of the organizations working in similar scientific and technical areas, is Chiron more likely to conclude an alliance with an organization with whom it has a joint patent or vice versais Chiron more likely to patent jointly with an organization with whom it has an alliance? To answer this question we must first identify the pool of organizations similar enough to Chiron that an alliance or a joint patent would not be out of the question. Our criterion here is that the organizations have a link to Chiron in two of the three databases used: alliance, patent or paper. There are 260 organizations that meet this criterion; these are our "pool." Even within this pool, joint patents and alliances are both rare events. Chiron has R&D alliances with 17 of these organizations and joint patents with 16. If alliances and joint patents were created randomly within this pool, the probability of a company having both an R&D alliance and a joint patent with Chiron would be: 16/260 * 17/260 = 0.004. The number of companies we might expect to see with both an R&D alliance and a joint patent would then be: 0.004 * 260 = 1.05. Instead there are six organizations with whom Chiron both patents and has R&D alliances. This suggests that Chiron chooses alliance partners preferentially from among those with whom it has joint patents or vice versa, that Chiron co-patents preferentially with those with whom it has an alliance. The relationship is not all that strong; after all, in the majority of cases alliance and joint patent partners differ.
Conducting the same analysis on the other bibliometric dimensions suggests no relationship to alliances in most cases as the number of overlapping organizations is equal to the number expected by chance. The one possible exception is organizations whose patents are cited by Chiron patents for whom we expect 8 overlapping organizations and find 12, or 1.5 times more than expected.
The possibility of some relationship between R&D alliances and joint patenting is intriguing. Previous work investigating Du Pont's joint patenting and other bibliometric indicators found that coassignees were not predictable using factors such as geography or industry which influence citation and coauthorship (Hicks, 2000). This made it seem more likely that behind each joint patent is a unique and substantive story of companies coming to share technology in spite of their natural distaste for doing so.
The lack of correlation between alliances and Chiron's bibliometric dimensions creates some obvious anomalies. Examining all alliances (not just R&D) we find that Chiron's top alliance partner is Johnson & Johnson with whom Chiron has joint ventures, marketing agreements, finance agreements and more complex arrangements. Apparently this relationship is all about marketing and not about technology because although Johnson & Johnson appears in many bibliometric dimensions, its rank is quite low. In contrast, Chiron has no agreements with American Home Products, arguably the company with which its technology appears most interdependent.
Although there may be some overlap between alliances and joint patenting, it seems prudent to consider what the bibliometric dimensions might be tracking other than strategic R&D alliances and whether they might be used as indicators that complement formal strategic alliances in enhancing our understanding of firm's scientific and technological networking.
We begin with coauthorship because here the clearest link to informal networking can be made. We maintain that co-authored papers tell us something about how knowledge moves. The movement of scientific and technical knowledge, particularly between universities and companies, is often called "technology transfer." Cohen et al. surveyed 511 university-industry research centers in the United States asking how effective were various technology transfer mechanisms. They conclude that:
Because papers are not effective in transferring technology, and this is often reported in the literature, papers can be seen as irrelevant to moving knowledge. However, using papers to study the patterns of knowledge distribution does not require that papers effectively convey knowledge. It would be enough if processes that transferred knowledge also tended to produce a paper. Then the papers would be signals of the underlying process and indicators of knowledge transfer could be developed using them. We argue that co-authored papers indicate links between firms and public sector research that effectively transfer technology. Indeed, of the five most effective mechanisms of transferring technology listed by Cohen et al., three are likely to produce co-authored papers: collaborative R&D, secondment to the university or secondment to the company.
Co-authored papers can be produced by other types network relationships through which science and technology can be linked. For example, a paper listing two addresses can result from a single author holding a joint appointment, indicating a substantive link between two institutions through which knowledge and expertise can be exchanged. Co-authored papers can result from students hired as they finish their PhD degrees (who list both their previous and current addresses when publishing their PhD research), indicating one of the most important mechanisms through which knowledge is diffused in the economy.
Of course, to expect a one-to-one matching between every informal relationship and a co-authored paper is unrealistic (Katz & Martin, 1996). In addition, collaborative relationships can be maintained over many years, but papers may appear in the SCI only once or twice; so the duration of a relationship is less reliably indicated by papers than its existence. Nevertheless, a set of co-authored papers can be used to construct indicators providing unique information. There is no other way to obtain a quantitative, longitudinal overview of informal linkages across all types of research-producing organizations.
Zucker and Darby have studied biotechnology in the US; they quote a manager as saying: "Copublishing is about as good an indicator as you can get of commonality of interests between [the company] and an academic collaborator. Although formal relationships are on a publicly available list, many relationships are not publicly acknowledged." They continue: "In this and other fieldwork we have repeatedly validated the usefulness of linking academic scientists to firms by bibliometric research on patterns of co-publication. ... this concept of linkage is powerfully predictive of firm success when academic star scientists are involved." (Zucker & Darby, 1995, p. 22).
Given the paucity of studies using coassigned patents, we can only speculate that it is likely that they too reflect an informal yet substantive level of technological networking.
The core of knowledge production is believed to rest in tacit knowledge, artifacts and the networks of communication through which these are developed and exchanged. Neither the objects nor the tacit knowledge in these chains can be communicated in a publication, so documents have often been dismissed as irrelevant to understanding the processes of research. Hope has been expressed that the links created through referencing might mirror the communications between authors of documents, but close study of referencing proved disappointing in this regard (Meyer, 2000; MacRoberts & MacRoberts, 1989). A way forward to interpreting the relationship between bibliometric indicators and networks of scientific and technical work has been provided by anthropologists.
Anthropologists of science have argued that the work of research connects heterogeneous elements. Latour and Woolgar pointed to the integration of informal communication with documents including the highly stylized scientific paper (Latour and Woolgar, 1989, p. 52-53). Hilgartner and Brandt-Rauf pointed to the chains of products from scientific work which brought together instrumentation, materials, craft skills, information, documents, informal communications and so on (Hilgartner and Brandt Rauf, 1994, p. 7). Because heterogeneous elements are integrated in these chains of products, a paper describing research points to other elements in the chain and so indicates that the authors possess certain tacit knowledge and materials. Papers carry signals about the areas in which researchers work, the craft skills they possess, the materials they use, their instrumentation and the quality of their work. Papers alert us to the existence of underlying tacit knowledge, skills, substances and so on possessed by the authors. Published papers thus point to unpublished resources. The same logic applies to the patent document.
Papers and patents also explicitly point to other papers and patents. The making of this link constitutes a suggestion that there may be some logical relationship between the resources underlying the linked documents. This hints that the documents' authors may have something to talk about; if they have actually communicated we would have a network type of link. Mowery, Oxley and Silverman (1998) found that alliances were more likely between firms with higher patent-to-patent citation interdependence. But this held only up to a certain point and that very highly interdependent firms did not form alliances (presumably because direct competitors often have very high patent citation overlap but direct competitors are unlikely to form alliances). If this result extends to the other dimensions, we may find that networks are more likely among those linked to some degree by referencing and citation than those not so linked or those highly linked.
Bibliometric documents and links have varying properties. Papers are research related; patents innovation related. Citations are incoming and references outgoing, suggesting a provide/use difference. Paper-paper links are scientific; patent-patent links technological, and paper-patent links join science and technology. The eight bibliometric dimensions exhibit the permutations of these properties which may enable a sensitive delineation of scientific and technological networks.
At this preliminary stage, we can only conclude that it might be possible to develop the 360 degree look at bibliometric citation dimensions into a tool to increase understanding of the networks of scientific and technological work. The referencing dimensions are a reflection of the somewhat unintentional and promiscuous linking processes of paper and patent referencing. The co-production dimensions reflect much more focussed, intentional, substantive investments of effort into joint work. None of the dimensions measures exactly same thing as formal alliances, suggesting that more informal processes underlie the bibliometric dimensions. In analogy with Von Hippel's concept of informal know-how trading, we might hypothesize that these processes of linkage are embedded in day-to-day knowledge work and yet are practically invisible at the strategic management level.
This has been a preliminary exploration of the relationship between R&D alliances and bibliometric indicators, with the emphasis on joint patenting. We have seen that joint patenting is very rare, though becoming more frequent. In the health technologies, public sector participation in patenting may explain the growth in joint patenting as universities and government laboratories carry over into patenting the collaborative instincts developed through publishing. Companies, on the other hand, seem to have a positive aversion to sharing intellectual property which they only rarely overcome, and it would be quite interesting to find out what prompts them to do so since it seems unlikely that a higher economic value to joint patenting is the motivation.
There may be some relationship between joint patenting and R&D alliances in the sense that partners for one are chosen preferentially from among those with whom the other relationship has been formed. Our dataset is however a small case study and only a much larger investigation could definitively establish this.
In general, the bibliometric indicators seem to track something slightly different from formal R&D alliances. We believe that they may track informal networking processes. The intentionally produced joint papers and patents should be the most direct indicators of this. Behind citations and references are the somewhat unintentional logical links made between organizations and so between the unpublishable resources underlying patents and papers. These may bear some relationship to potentials for networking.
None of the dimensions is likely to correlate with management's view of its networks in the same way as formal alliances do. Even the intentional dimensions of coauthorship and copatenting are probably intentional only on the part of the researchers involved and only dimly visible to upper management. We would argue that this is a strength of the bibliometric indicators, getting closer to the day-to-day work of research and development. Whether there is more substance in the formal alliances or in the bibliometric dimensions remains an open question.
 Foreign coassigned patents are more numerous. Hagedoorn has investigated this more closely and finds that this is accounted for by intra-keiretsu joint patenting by Japanese companies (Hagedoorn, private communication).
 Health technologies are: biotechnology, pharmaceuticals, medical equipment and medical electronics.
 Following Wouters, we will distinguish references from citations. References are "outgoing" and citations are "incoming." That is, references are in lists at the end of scientific papers, and citations are found in the Science Citation Index indexed under the receiving paper. The contents of references are bibliographic descriptions of the receiving papers, and the contents of citations are abbreviated bibliographic descriptions of the papers giving the citation.