ABSTRACT
This paper explores the evolutionary trajectories of cluster research, building upon the sociology of science concept of the invisible college, and it undertakes a core–periphery analysis of the literature. We build a database that includes 8,381 articles, collected from Web of Science, that cite the foundational works of cluster research, and we perform a longitudinal analysis of its evolution from 1985 onward, identifying the core and periphery, in terms of keywords and concepts, for each period (six-year window). We find evidence that cluster research has a core–periphery structure. Literature develops thanks to new inputs from the periphery, which increases over time as the core progressively shrinks. The periphery becomes fragmented and is characterised by subgroups of small communities. Drawing on the metaphor of the invisible college, we argue that this evolutionary trajectory is not exclusive of the cluster but might possibly characterise other scientific concepts.
JEL CODES:
Acknowledgments
This work greatly benefited from comments received from participants at the AAG conference in Los Angeles (2013); Florence University Workshop (2016); Geoinno conference in Barcelona (2018); Rethinking Clusters Workshop in Florence (2018). Special thanks go to Robert Hassink for precious comments on an earlier version of this work.
Disclosure statement
No potential conflict of interest was reported by the authors.
Supplementary material
Supplementary data for this article can be accessed here
Notes
1 For more details on conceptual stretching, see Sartori (1970), Van Meeteren et al. (Citation2016), and Fröhlich and Hassink (2018).
2 We chose 1985 as the starting year because this is when the systematic collection of articles in the WoS database that we have considered began. We concluded our observation in 2013, because our empirical research took place in 2014.
3 Given that the list of foundational works also included book chapters or books that would be omitted by a direct search on WoS, which focuses on journals, we also performed an indirect search for articles that we knew would have included the books we were looking for in their reference lists. Then, clicking on the references, we were able to retrieve a list of WoS articles that cited them.
4 The literature also provides many looser definitions of core–periphery, which can be used to describe a wide number of real-world networks. For example, according to Borgatti and Everett (Citation1999, 375), a core–periphery network is characterised by the presence of a cohesive subgroup of core actors and a set of peripheral actors that are only loosely connected to the core.
5 The fit function is the correlation between the observed network and a matrix consisting of values of 1 in the core block interactions and values of 0 in the peripheral block interactions. The breakdown between the core and the periphery was done in such a way as to maximise the correlation. To test the robustness of the solution, we ran the algorithm a number of times from different starting configurations.
6 The algorithm, which uses a heuristic method that is based on modularity optimisation, allows for different resolution parameters. The default is the standard Louvain method (resolution = 1); however, higher resolutions produce a larger number of clusters, while lower resolutions produce a lower number of clusters. Given that the results of the algorithm are influenced by the (random) choice of the starting point, we ran the algorithm numerous times, with several restarts, to select the best partition from all the restarts. By comparing the partitions obtained with the same resolution parameter through the Cramer’s V, Rajski, Adjusted Rand Index, we identified the ‘optimal’ number of clusters.
7 This latter cluster included articles that attempted to explain the basis of the innovative capacity of clusters and districts, starting with the institutions that characterised them.