Network effects are critical for research collaborations
At least since endogenous growth theory, knowledge and ideas have been considered key contributing factors to economic growth (Romer 1990). One important channel through which knowledge is created and diffused is the publication of research papers, which are increasingly produced via collaborations across co-authoring researchers, in economics as in other fields (Wuchty et al. 2007, Henriksen 2016, Kuld and O’Hagan 2017).1 Teaming up allows researchers to exploit synergies and economies of scale by pooling ideas, skills, time, and funds. While there is no definitive consensus on the effect of co-authorship on research output, recent studies point to a positive relation between co-authorship and the number and quality of papers produced (Ductor et al. 2014, Ductor 2015, Sommer and Wohlrabe 2017).
The common wisdom is that co-authorship networks play a crucial role in both knowledge creation (as multiple authors combine their ideas to come up with something new) as well as in knowledge diffusion (researchers learn from each other and can pass on the newly acquired knowledge to their existing and future contacts). Indeed, the empirical evidence suggests that knowledge spills over directly to co-authors (Azoulay et al. 2010, Borjas and Doran 2015), as well as indirectly to the broader network of collaborating researchers (Hsieh et al. 2018).
In our work (Essers et al. 2020) we study how research collaborations in economics are formed and maintained, focusing on the role of the co-authorship network structure.2 Departing from earlier literature based on research collaborations observed in publications across peer-reviewed economic journals (Fafchamps et al. 2010), we draw on the full co-authorship network of the IMF working papers series and construct a novel dataset covering almost three decades (1990-2017) of publications.3 The sample of IMF working papers provides several advantages. The authors of these papers, predominantly IMF staff, are arguably less exposed to the ‘publish or perish’ conditions and other strategic motives (e.g. career advancement) present in academia. The sample thus provides an ideal testing ground to examine the endogenous nature of co-authorship formation. Also, as it is based on working papers rather than peer-reviewed articles, it suffers less from the problem of long publication lags. Finally, we take advantage of the richer demographic and employment information on authors to better identify the importance of network effects.
A novel dataset
Similar to the trends observed in economics journals, the number of IMF working papers’ authors has increased at a faster rate than the number of publications since 2010, pointing to a growing number of co-authors for each paper (Figure 1). The distribution of the number of papers written by any given author, however, is highly right-skewed, with the large majority of authors having published only one paper and very few producing more than 20. A larger number of authors per paper is also associated with more citations (Figure 2), likely reflecting synergies across co-authors. Within the IMF, the Research Department accounts for the most papers in absolute numbers and for the largest share of publishing economists by department; its papers also tend to get more citations.
IMF research authorship has become more diverse, reflecting changes in the IMF workforce, i.e. women and people with non-US and non-European backgrounds increased their shares in total working paper output. In terms of topic coverage, unsurprisingly, most working papers deal with macroeconomics, international economics, financial economics, and public economics. However, in recent years, increased attention has been dedicated to less conventional or newer topics for the IMF, including climate change and gender.
Figure 1 Publications and authors
Figure 2 Annualised citations per number of authors
The co-authorship network
A closer look at the co-authorship network of IMF working papers reveals that IMF staff dominates in terms of the number of authors and working papers, followed at a clear distance by academics, central bankers, and staff from other international organisations (Figure 3). Authors affiliated with government agencies, private companies, and non-profits occupy a much less prominent role in the network. IMF staff has been collaborating with researchers from many different organisations within the same institutional category, but at the internal margin these collaborations are very heterogeneous. The most recurrent links are with researchers from the World Bank and US and UK universities, and, in second instance, from other European universities and advanced economy central banks. Among the institutions involved in at least ten papers, there are only a few from emerging market or developing economies and no government agencies or non-profits (Figure 4). Also, research collaborations within the same IMF department are more common than co-authorship across different departments or with external institutions.
Network statistics, including the size of the giant component and the average degree, distance, and clustering coefficient, confirm that the co-authorship network has not only become larger over time but also more integrated. Taken together, our findings indicate that the IMF working papers network exhibits ‘small world’ properties (Newman 2001, Goyal et al. 2006). That is, the network consists of a large set of authors with typically few direct co-authors but the majority of authors are nonetheless in some way connected to each other, mostly through short chains of co-authors. Further, the degree of overlap in co-authorship is very high. As Goyal et al. (2006) and Hsieh et al. (2018) show, similar patterns are also found in the network of economics journals and derive from a network architecture of interlinked ‘star’ authors, prolific researchers that act as connectors of different clustered parts of the network.
Figure 3 Core author-level network
Notes: The figure illustrates the core author-level network based on IMF working papers co-authorship over 1990–2017. The sample includes authors who contributed to at least ten publications and collaborated with at least ten different co-authors. The nodes represent authors and edges represent publication co-authorship. The node size is proportional to the number of publications by an author (including single-authored papers). The edge width is proportional to the number of collaborations between two authors. The node colours are according to the modes of authors’ affiliation category: light blue denotes the IMF, pink denotes other international organizations, red denotes universities, light green denotes central banks, dark blue denotes private organizations.
Figure 4 Core institution-level network
Notes: The figure illustrates the core institution-level network based on IMF working papers co-authorship over 1997–2017. The sample includes institutions that contributed to at least ten publications. The nodes represent institutions and the edges represent co-authorship by authors affiliated with the respective institutions. The node size is proportional to the number of publications to which at least one author affiliated with the institution contributed. The edge width is proportional to the number of collaborations between the two respective institutions. The node colours are according to institutional category: light blue denotes the IMF, pink denotes other international organizations, red denotes universities, light green denotes central banks, dark blue denotes private organizations.
Determinants of research collaborations
To assess the importance of network effects for starting and continuing knowledge generation, we empirically investigate the determinants of initial and subsequent research collaborations in our sample of IMF working papers. Methodologically, our analysis is closest to the work of Fafchamps et al. (2010), who study the determinants of co-authorship in a large sample of peer-reviewed articles in economics journals indexed in the EconLit database. In addition to the characteristics analysed in Fafchamps et al. (2010), we also look at the authors’ gender, region of citizenship, and the IMF department in which they are employed. This extra information allows us to test additional hypotheses and, importantly, to control for possible confounding factors when making inference on the effect of pure network variables on the probability of collaborating.
We find that, above and beyond other pair and individual author characteristics, two researchers are significantly more likely to collaborate, especially in the case of first-time collaborations, if they are closer to each other in the existing co-authorship network. This result is in line with the interpretation that collaboration networks may transmit important, and not directly observable information about authors (the so-called network effects) that helps to overcome matching frictions. Once this information is internalised through actual collaboration, the network effects strongly diminish. In addition, we find that more productive authors (with more well-cited papers) are more likely to start and continue collaborations. At the same time, researchers with an established network of co-authors seem to be less in search of new ones. Yet, authors with different co-author network sizes are more likely to start and maintain a collaboration, possibly reflecting collaborations between senior staff members (generally tasked with more managerial duties) and junior staff (who have more time for research). Finally, our results indicate that being employed in the same department and having citizenship of the same region help to start and repeat collaborations, while greater overlap in broadly defined research areas does not.
We conclude that incentives aimed at connecting researchers (especially those that are initially more distant from each other in the overall network) and spurring collaborations across them are likely to result in more dynamic knowledge generation, which in turn should promote economic growth.
Authors’ note: The views expressed in this column are those of the authors and do not necessarily represent those of the IMF, its Executive Board, its management, the National Bank of Belgium, or the Eurosystem.
Adams, J (2012), “The Rise of Research Networks”, Nature 490: 335–336.
Azoulay, P, J S Graff Zivin and J Wang (2010), “Superstar Extinction”, Quarterly Journal of Economics 125(2): 549–589.
Borjas, G J and K B Doran (2015), “Which Peers Matter? The Relative Impacts of Collaborators, Colleagues, and Competitors”, Review of Economics and Statistics 97(5): 1104–1117.
Ductor, L (2015), “Does Co-authorship Lead to Higher Academic Productivity”, Oxford Bulletin of Economics and Statistics 77(3): 385–407.
Ductor, L, M Fafchamps, S Goyal and M J van der Leij (2014), “Social Networks and Research Output”, Review of Economics and Statistics 96(5): 936–948.
Ductor, L, S Goyal and A Prummer (2018), “Gender and collaboration in economic research”, VoxEU.org, 30 April.
Essers, D, F Grigoli and E Pugacheva (2020), “Network Effects and Research Collaborations“, IMF Working Paper 20/144.
Fafchamps, M, M J van der Leij and S Goyal (2010), “Matching and Network Effects”, Journal of the European Economic Association 8(1): 203–231.
Georg, C-P and M E Rose (2016), “A ranking of financial economists by centrality”, VoxEU.org, 16 January.
Goyal, S, M J van der Leij and J L Moraga-González (2006), “Economics: An Emerging Small World”. Journal of Political Economy 114(2): 403–412.
Henriksen, D (2016), “The Rise in Co-authorship in the Social Sciences (1980-2013)”, Scientometrics 107(2): 455–476.
Hsieh, C-S, M Konig, X Liu and C Zimmermann (2018), “How researcher rankings and research funding instruments can gain from information about co-authorship networks”, VoxEU.org, 26 November.
Iaria, A, C Schwarz and F Waldinger (2018), “The importance of frontier knowledge for the generation of ideas”, VoxEU.org, 26 January.
IEO (2011), Research at the IMF: Relevance and Utilization, Evaluation Report.
Kuld, L and J O’Hagan (2017), “The trend of increasing co-authorship in economics: New evidence”, VoxEU.org, 16 December.
Newman, M E J (2001), “The Structure of Scientific Collaboration Networks”, PNAS 98(2): 404–409.
Sommer, V and K Wohlrabe (2017), “Citations, Journal Ranking and Multiple Authorships Reconsidered: Evidence from Almost One Million Articles”, Applied Economics Letters 24(11): 809–814.
Wuchty, S, B F Jones and B Uzzi (2007), “The Increasing Dominance of Teams in Production of Knowledge”, Science 316(5827): 1036–1039.
1 Throughout the column we will use co-authorship as a proxy for research collaboration. As Adams (2012, p.335) notes, since few researchers would casually surrender credit for their publications, one can assume that sharing authorship generally reflects a tangible engagement. Besides co-authorship, research papers embed other forms of (informal) collaborations, as captured by citations of other papers (Iaria et al. 2018) and acknowledgements (Georg and Rose 2016), but these do not necessarily entail the same level of engagement.
2 Ductor et al. (2018) use co-authorship network data to examine gender differences in research collaborations.
3 The IMF working papers series is one of the main research outlets of the institution. Officials in central banks, government entities, academia, and think tanks report using IMF working papers frequently as a reference in policy discussions. At the IMF itself they are widely read by staff and serve as ‘a vehicle to disseminate emerging ideas within the institution, to share new types of analysis and new ways of looking at country policies (IEO 2011).