A new approach to analyzing patterns of collaboration in co-authorship networks: mesoscopic analysis and interpretation

Abstract

This paper focuses on methods to study patterns of collaboration in co-authorship networks at the mesoscopic level. We combine qualitative methods (participant interviews) with quantitative methods (network analysis) and demonstrate the application and value of our approach in a case study comparing three research fields in chemistry. A mesoscopic level of analysis means that in addition to the basic analytic unit of the individual researcher as node in a co-author network, we base our analysis on the observed modular structure of co-author networks. We interpret the clustering of authors into groups as bibliometric footprints of the basic collective units of knowledge production in a research specialty. We find two types of coauthor-linking patterns between author clusters that we interpret as representing two different forms of cooperative behavior, transfer-type connections due to career migrations or one-off services rendered, and stronger, dedicated inter-group collaboration. Hence the generic coauthor network of a research specialty can be understood as the overlay of two distinct types of cooperative networks between groups of authors publishing in a research specialty. We show how our analytic approach exposes field specific differences in the social organization of research.

Appendices

Appendix 1: Data

Guided by our informants we developed lexical queries to retrieve from Web of Science the literature covering a specific speciality—sometimes a single term, sometimes a 2–3 line long query using Boolean operators. Our informants validated the data sets, checking whether crucial authors whom they expected to see were actually represented, and whether out-of-scope topics had been accidentally introduced. The data was then further processed to eliminate articles with only one author or unknown authors (indicated by ‘Anon’ in the WoS records), or with only a single reference, as the latter records typically do not refer to original research articles or reviews.

This study is part of a larger dissertation research project. To ensure the anonymity of the human participants of the field study part of this research in accordance with the IRB⁶ approved protocol for the dissertation research we cannot disclose the lexical queries that would provide access to the exact data sets discussed in this report. If access to the networks build from the data used in this study is needed for a particular research or validation purpose we are happy to work out a solution that retains the anonymity of our study participants.

Data field A

Time span: 1991–2008, number of records: 53,947 (retrieved on 22 Jun 2009)

Data field B

Time span: 1991–2008, number of records: 12,817 (retrieved on 14 Nov 2008)

Data field C

Time span: 1987–2008, number of records: 30,636 (retrieved on 17 Dec 2008).

We derive the geographical affiliation of a cluster at continent level from the country affiliation listed in the Web of Science record for each publication. We represent each cluster by all the publications any of its authors has been a co-author of. Then we determine the country affiliation that is most often listed for papers published by cluster authors. In cases where the second placed country is listed at least 50% as many times as the most often listed country, and if these two countries belong to different continents, we assign a mixed, two continent geographical affiliation

Appendix 2: Methods

In Guimera et al. (2007) each node of a clustered network was assigned to one out of seven node roles. The seven node roles are defined by the values of two parameters that quantify how a node is connected to the other nodes in its cluster and how the links going outside the cluster are distributed.

The first parameter, z, is simply the inside-its-cluster-degree of a node, normalized by subtracting the average inside-the-cluster degree and then dividing the result by the standard deviation. Nodes that have z < 2.5 were called non-hubs, while nodes with z ≥ 2.5 were called hubs. In other words, hubs are well-connected inside their own cluster with a normalized threshold beyond 2.5 (z > 2.5).

The second parameter is the participation coefficient P that quantifies how a node distributes its outside links among the clusters. For a node i in a network of N clusters it is defined as follows:

$$ P_{i} = 1 - \sum_{\text{s}} \left( {k_{\text{s}}^{i} /k^{i} } \right) $$

where s = (1…N), k ⁱ_s  = number of links of node i to nodes in cluster s, k ⁱ = the total degree of node i.

From this definition it follows that P is normalized for each node so that the value is within 0 and 1. A node that connects to only one cluster would have a small P whereas a node that connects to a lot of clusters would have a high P.

Non-hubs are classified into 4 types based on their P values.

• P ≤ 0.05: Ultra-peripheral (R1) with its connections restricted to its own cluster

• 0.05 < P ≤ 0.62: Peripheral (R2) with limited outside connectivity

• 0.62 < P ≤ 0.8: Satellite connector (R3) with good connectivity with a number of outside clusters

• 0.8 < P: Kinless (R4) with outside connectivity distributed evenly among all clusters

Similarly hubs are categorized into 3 types based on their P values.

• P ≤ 0.30: Provincial hub (R5) that is well-connected only inside its own cluster

• 0.30 < P ≤ 0.75: Connector hub (R6) that connects to many clusters

• 0.75 < P: Global hub (R7) that connects to most clusters

We have used the same set of threshold values to assign roles to nodes.