Social network analysis of international scientific collaboration on psychiatry research

Achievements in scientific research are published in the form of papers and the status of co-authorship in papers reflects collaboration among authors. M.smith was one of the scientists who studied the growth of co-authorship papers made by multi-author and viewed co-authorship of papers as a importance scientometrics indicator of researching on collaboration among authors [18].

In order to show the main co-authorship structure of the network, we selected the top 100 prolific authors during 2003 to 2012 in this study. This threshold resulted in the top 100 prolific authors who must publish 43 co-authorship papers. (Three authors who have not cooperated with other authors were deleted), so the Figure 1 was a co-authorship map made up of the top 100 authors visualizing the structure of authors’ collaboration network. The map was composed of four sub-networks which are not connected with each other. The line value and the distance between two vertices represent the collaborative strength, while thickness of the line represents the number of co-auhorship papers. In this authors’ collaboration network, the highest degree centrality of Pine, DS was 145 indicated he had 145 collaborators, so he was the most key author of the co-authorship network. Seidman, LJ had got the highest betweeness centrality which indicated that he possessed and controlled a great deal of research resource. Fava, M had the lowest closeness centrality which indicated he was in a core position of the whole network (see Table 2). In collaborative network, betweeness centrality reflects the author’s function. The lack of author with the highest betweeness centrality lead to connection interruption of collaborative network. That Seidman, LJ had got the highest betweeness centrality indicated he had the power to control collaborative relationship. In collaborative network, the closer one author is to the other author, more easily are information communication and research collaboration. Freedman, R had the lowest closeness centrality which indicated that he possessed and controlled a great deal of research and was in a core position of the whole network (see Table 3).

Hierarchical clustering usually categorizes prolific authors and creates a hierarchy of clusters which can be represented in a tree structure called a dendrogram. Through Hierarchical clustering analysis, we got Figure 2 and we divided 97 authors into 3 sub-networks (see Figure 1). The largest sub-network included 60 nodes and 754 lines. In this subnetwork, the average path length was 2.091 and the average clustering coefficient was 0.651 which indicated obvious clustering effect and characteristics of small world. In this subnetwork, Freedman, R had the highest centrality degree and his research direction was mainly child and adolescent psychiatry. Siever, LJ had the second centrality degree and his research direction was mainly pathophysiology of child and adolescent schizophrenia. Carter, CS had the third centrality degree and his research direction was mainly cognitive dysfunction of schizophrenia. The second sub-network included 35 nodes and 166 lines. In this sub-network, the average path length was 2.545 and the average clustering coefficient was 0.502 which indicated obvious clustering effect and characteristics of small world. In this subnetwork, Seidman, LJ had the highest centrality degree and his research direction was mainly brain imaging of schizophrenia. Liberman, JA had the second centrality degree and his research direction was mainly on drug treatment of schizophrenia. Sharma,T had the third centrality degree and his research direction also mainly on drug treatment of schizophrenia. The third sub-network only included Egan, MF and Kendler, KS and their research direction was molecular genetics of psychiatry.

A K-plex is a maximal sub-network in which each node has at least connected with other nodes except these K nodes directly within the sub-network. It is widely used in collaborative network which is undirected and has more value network. In the first, determine the condensation degree of subgroup. The critical value ‘C’ is bigger, the condensation degree of subgroup is stronger. If the value between ‘g’ nodes to ‘g-k’ nodes all at least not less than ‘c’ in a subgroup, we called this subgroup ‘c’level K-plex [19]. In order to exclude the phenomenon of the fewer number of collaboration, ‘C’ was determined to be ‘4’. It indicated the authors who collaborated with others less than 4 time would no longer appear in K-plex and these subgroups were higher cohesive in which the members were keeping a relatively close relationship. By using UCINET there were totally 776 ‘K-2’ K-plexs which collaborative frequency was greater than 4. It indicated the number of collaboration between any author in ‘K-2’ K-plex and other two authors was no less than 4 time. Pine, DS appeared in 449 ‘K-2’ K-plexs and Weinberger, DR appeared in 319 ‘K-2’ K-plexs.

Since the 1920s,with the rapid development on the scale and scope of research collaboration, the collaborative papers among institutions increased 46% [20]. Analysis on relationship network of academic institutions in research collaboration is of great significant to research mechanism, influence factors and academic information exchange model in scientific collaboration. There were 19475 papers which belonged to inter-institution collaboration among 36557 papers from SCI during 2003 to 2012. The number of papers has increased from 1072 in 2003 to 1598 in 2012. These papers covered 5182 actual institutions and the appearing frequency of institutions was totally 36653.The largest collaboration in our sample involved 47 institutions. Seen from Table 4 which described the annual change in institutions, the appearing frequency of institutions grew significantly in 2005 and 2009 and the number of actual institutions increased in the two years while the achievements in scientific research rose respectively in 2005 and 2009. It suggested that the scale of collaboration was related with the output of scientific research positively. We selected the top 100 institutions with appearing frequencies more than 57 to form a map visualizing the structure of institutions’ collaboration network in the field of psychiatry during 2003 to 2012 (see Figure 3). The size of node represents centrality in collaborative network. Harvard University had the highest degree centrality, and Yale University had the highest betweenness centrality and the lowest closeness centrality (see Table 5). It shows Harvard and Yale University were in high level of collaboration. The distance and thickness of the line between two nodes represent their collaborative strength and the number of collaborative papers respectively. From Figure 3, National Institutes of Health (NIH), Havard University and Yale University were in the center of collaborative network which had a fundamental impact on the development of psychiatry; while University of Zurich and University of Louisville were on the edge of collaborative network. Relative to the institutions in the center, scientific research strength of institutions on the edge were slightly inferior and collaboration closely among the institutions in the center reflected obvious ‘center effect’ in the process of co-authorship while the institutions on the edge collaborated looser. Analysis on Core-Periphery is the quantitative study of various network. We applied this method to collaborative network and used UCINET 6.0 to calculate ‘correlative value of collaborative network’ being 0.628. From Figure 4, we detected the polarized Core-Periphery structure was very obvious. In other words, collaborative network showed obvious regional characteristics and firstly select research institution in the close geographical position. That the highest collaborative frequency of Harvard University and Yale University indicated these two research institutions have a close research collaborative relationship.

Studies showed that research papers produced by international collaboration had larger impact [21]. In the 1990s, collaborative papers among countries increased by 115% within the amount of research paper from SCI [20]. During 2003 to 2012, the total number of countries was 107. From Figure 5, the countries with highest productivity include USA, England and Canada. There were 12808 papers in psychiatry research originated from USA than other countries and 60.6% of the total number of documents. There were 99 papers in China, ranked the 19th place. We chose the top 30 countries with appearing frequencies more than 86. Figure 6 was the map of the scientific collaboration of the most productive countries in the world. The network included 30 nodes and 1032 lines. In this network, the average path length was 1.076 and the average clustering coefficient was 37.572 which indicated obvious clustering effect and characteristics of small world. We applied Core-Periphery analysis and used UCINET 6.0 to calculate correlative value of collaborative network being 0.899. From Figure 7, we detected the polarized Core-Periphery structure was very obvious.

We analyzed international scientific collaborative effect on national scientific through the correlation between the number of corresponding nodes and scientific research achievements. The national names, the number of ties and the product of papers which the 30 round nodes corresponded with were listed in the Table 6. We found that there were 21106 papers produced by 30 countries which collaborated more frequently accounting for 58% of the total output in 107 countries. From the Table 6, that the international order of these 30 countries was identical with their research output showed that international scientific collaboration had great influence on output of scientific research in this field. Scientific collaboration was basically correlated with the output of papers positively and the countries which had frequent scientific collaboration had larger research output.

From centrality analysis (see the Table 7). That the highest degree centrality of USA was 5783 and the highest betweenness centrality of USA was 1.621, while the lowest closeness degree of USA was 29 showed USA was the center of international scientific collaboration network in psychiatry field in the world. The large number of research output maked USA a major producer of international publications.