Bodong Chen

Crisscross Landscapes

Notes: Wagner, C. S., & Leydesdorff, L. (2005). Network structure, self-organization, and the growth of international collaboration in science. Research Policy, 34(10), 1608–1618. https://doi.org/10.1016/j.respol.2005.08.002

2017-08-31


References

Citekey: @Wagner2005-ib

Wagner, C. S., & Leydesdorff, L. (2005). Network structure, self-organization, and the growth of international collaboration in science. Research Policy, 34(10), 1608–1618. https://doi.org/10.1016/j.respol.2005.08.002

Notes

Summarize: Very interesting study of international collaboration using the sociological concept of preferential attachment to make an argument about the self-organizing feature of international collaboration (driven by all sorts of rewards). Two references regarding different types of collaborators highlighted near the end are also quite interesting and could apply to an analysis of an emerging, interdisciplinary area like learning analytics and MOOCs.

Highlights

Persson et al. (2004) show that cita- tions to articles resulting from international collabo- rations grew faster than those referring to domestic collaborations. Narin (1991) shows that internation- ally co-authored articles are more highly cited. Despite this body of evidence, the question of why this class of research is growing or why it has a relatively high impact needs more discussion (Katz and Hicks, 1997; Wagner-D ̈obler, 2001). (p. 1)

Physi- cists studying network dynamics have found interesting regularities within the networks created by scientific co-authorships. In the process, they have revealed the mechanism of preferential attachment as a structur- ing factor in scientific collaboration. Using the tools developed by Barab ́asi and Albert (1998), Newman (2001, 2004), and Barab ́asi et al. (2002) we investigate whether these mechanisms also apply at the interna- tional level and what findings might mean for research policy. (p. 2)

Our data suggest that a centre–periphery model may operate at a regional level, but at the global level, that model does not explain the dynamic revealed in the network data. Indeed, various scientific centres both collaborate among and compete with one another for partners at the international level. (p. 2)

  1. A brief review of the literature examining the growth of international collaboration (p. 2)

Over the past 25 years, a number of reasons have been suggested to explain the growth of ICS. These can be divided into factors that are considered either internal or external to science. Table 1 cross-references these factors (p. 2)

Nevertheless, they do not explain the system that emerged between 1990 and 2000 (see data we presented in a separate paper; Wagner and Leydes- dorff, 2005) which shows significant shifts in network structures within that decade (p. 2)

  1. Applying network theory as an explanatory factor (p. 3)

Similarly, the cost of “megascience” as a motiva- tor for international collaboration can be inferred from specific activities such as growth of research around the Human Genome Project, but the growth of ICS is not limited to these large-scale projects. Megascience cannot be shown to be the sole or even the primary driving factor behind the growth of ICS between 1990 and 2000. (p. 3)

We developed another hypothesis: ICS can be viewed as an emergent, self-organizing system where the selection of a partner and the location of the research rely upon choices made by the researchers themselves rather than emerging through national or institutional incentives or constraints. (p. 3)

The concept of self-organisation seemed particu- larly relevant in this case because, as we discussed above, no single factor can explain the rise of ICS. No centralized authority guides the organization of inter- national science; no single driver has clear explana- tory power. (p. 3)

  1. Data and methods (p. 4)

Newman (2000a,b, 2001, 2004) has developed tools to reveal the structural characteristics of scientific col- laborations. He has shown that collaborative scientific networks have surprisingly short node-to-node dis- tances (with a node being a single author) and high clustering coefficients (2000). (p. 4)

4.1. Data (p. 4)

Data were drawn from the Science Citation Index (SCI) CD-Rom version 2000 of the Institute for Sci- entific Information (ISI) for six fields of science. (p. 4)

For the relevant journals, author names and addresses appearing in 19,147 articles were drawn from 65 journals covering the six fields. Author names were taken into the database as recorded; no attempt was made to adjust for spelling variants. (p. 4)

Our data was limited to 1 year – co-authorships in 2000 – so we are not able to show the dynamics of evolving networks. By comparing across fields in 1 single year – that is, in a static design – we seek to use these tools to test for preferential attachment as a structural phenomenon at the global level across fields. We searched for the specific log–log distribution when a degree-based measure is applied to data on co- authorship events in the six cases for the year 2000. (p. 5)

In this study, our objectives are more modest. We wish to show that the data exhibit the expected forms of power-law distributions with fat tails and hooks to test for preferential attachment; thus, we seek to use the fits only qualitatively as an indicator for this dynamic. (p. 5)

A scale-free network is characterized by the following scaling behaviour in P( k): P(k) ∼ k− γ where γ is the scale exponent. (p. 5)

4.3. Fitting the power-law distribution (p. 5)

In order to determine the exponent of a power-law, the data can be fit on a log–log scale. The exponent can then be determined as the coefficient of the line through the data with a best fit. (p. 5)

  1. Results (p. 5)

The network structure and the co-authorship dis- tribution exposed in our research suggest that, at the international level, networks of co-authorship display a preferential attachment mechanism similar to that found in studies of scientific co-authorship networks. (p. 5)

In findings similar to others (Barab ́asi and Albert, 1998; Albert and Barab ́asi, 2000; Barab ́asi et al., 2002) the networks of collaborations shown in the figures below have fat- tailed degree distributions. This has been interpreted as a power-law form (Barab ́asi and Albert, 1998). (p. 6)

The cost of adding a new link is a limiting factor in any social network, par- ticularly one focused on collaboration. (p. 7)

  1. A sociological explanation (p. 7)

It appears that preferential attachment operates in the middle part of the distribution shown in the figures. We submit that both the hook and the tail of the distributions can be considered as the institutional constraints on the dynamics. The hook can be identified with the arrival of newcomers into the field. This is a condition for the institutional reproduction. The tail can be consid- ered as representing an elite group of scientists who no longer compete for more prestigious co-authorship relations, some of whom are “terminants” at the end of their career. (p. 7)

These data suggest that the scale-free distribution of co-authorships that Newman (2001), Jeong et al. (2001), and Barab ́asi et al. (2002) found for fields of science in general can also be retrieved for interna- tional collaborations using a static analysis (for a single year). (p. 7)

To support this explanation, we turn to the work of Braun et al. (2001), following Price and G ̈ursey (1976), on the role of various actors within co-authorship net- works. Braun et al. built upon the work of Price and G ̈ursey to explore the dynamics of links within co- authoring communities. Price and G ̈ursey (1976) iden- tified four types of co-authors in a review of collab- orations over a 10-year period: terminants at the end of their career, newcomers at the beginning of their career, transients perhaps from another field who pub- lish in this field only once (this could also be considered as “new entrants” per Jeong et al. 2001), and continu- ants who publish a large number of papers in the year preceding and the year following the analysis. (p. 7)

Amaral et al. (2000) also finds a deviation from a power-law in some small world networks. They suggest that preferential attachment may be hindered by two classes of factors: (1) aging of the vertices and (2) the cost of adding links to the vertices. (p. 7)

skewed: continuants are likely to co-author and rarely appear as single authors; the overwhelming number of papers is co-authored by continuants. They report that “co-authorship relations among these three categories [newcomers, transients, and terminators] are usually also mediated by continuants” (p. 508). (p. 8)

However, some become so well connected that they no longer com- pete for reputation and therefore appear at the fat tail end of the distribution: they compete to build networks of intellectual followers of the next generation. (p. 8)

The more senior (or reputed) members of a group hold privileged social and technical information (Meyerowitz, 1985). Newcomers and transients seek access to this information as well as recognition within their field. As a result, the continuants (and perhaps the terminants), acting as hubs within their scientific networks, are attractive collaborators. (p. 8)

In general, the con- tinuants and, in some cases terminants, act as gate- keepers to newer entrants into the network, creating a social dynamic within the network. These dynamics of institutional reproduction distort the self-organising (p. 8)

features of the network, resulting in a less-than-perfect fit to the power-law form. (p. 9)

  1. Theoretical implications of this research This article argues that international collaboration in science at the sub-field level can be shown to self-organize based on rules of preferential attach- ment with social constraints. The networks examined here have self-organizing features, suggesting that the spectacular growth in international collaborations may be due more to the dynamics created by the self- interests of individual scientists rather than to other structural, institutional or policy-related factors that have been suggested by others. (p. 9)

The many individ- ual choices of scientists to collaborate may be moti- vated by reward structures within science where co- authorships, citations and other forms of professional recognition lead to additional work and reputation in a virtuous circle. Highly visible and productive researchers, able to choose, work with those who are more likely to enhance their productivity and cred- ibility. These “continuants” mediate the entrance of juniors into this network. This creates a competition for collaborators. (p. 9)

  1. Policy implications of this research (p. 9)

Other research suggests that knowledge creation may result from unstable networks with weak ties, where churn exposes researchers to new ideas and methods (Granovetter, 1973; Cowan and Jonard, 2003). (p. 9)

The network of international collaboration is highly dynamic, quickly changing, and very influential. As such, it feeds back into the national, regional, and local levels, influencing the organization of science. This dynamic organization presents particular challenges to national S&T policymaking. The first challenge is ensuring that knowledge flows to critical places within the research system. (p. 9)

To the extent that networks increase efficiency by connecting rather than recreating capabilities in each national system, the overall system may benefit, but the national systems may lose out, at least in the near term. (p. 9)

Braun, T., Gl ̈anzel, W., Schubert, A., 2001. Publication and cooper- ation patterns of the authors of neuroscience journals. Sciento- metrics 51, 499–510. (p. 10)

Price, D., G ̈ursey, S., 1976. Studies in scientometrics. Part 1. Tran- sience and continuance in scientific authorship. In: International Forum on Information and Documentation, pp. 17–24. (p. 11)