Unequal International Research Collaboration Networks of U.S. Universities and Foreign Countries in the Top Journals

• In Event: Comparative education and international perspectives

Proposal

Introduction
As knowledge production expanded, traditionally non-research universities (hereafter TNRs) in the U.S. significantly increased scientific activities over time (Shamekhi et al., 2019; Yu, 2022). With this trend, TNRs enhanced international collaborations (Yu, 2022). However, it is unclear whether TNRs published papers in highly-cited journals, which are often regarded as high-quality journals (Glänzel & de Lange, 2002; Kim et al., 2010), as they increase international collaborations. Because research universities were the dominant actors in the international research collaboration networks in the U.S. (Yu, 2022), TNRs would have difficulty in publishing papers in highly-cited journals. If this is true, the current international knowledge production system is discriminatory to TNRs. Therefore, this study aims at analyzing international research collaboration networks of U.S. universities in highly-cited journals.
Theoretical Framework
The Matthew effect (Merton, 1968) guides this study (Yu, 2022). As new scientists choose highly productive authors as their collaborators, famous researchers boost research production and the gap between renowned and new researchers widens (Barabási et al., 2002; Marginson, 2021). Universities showed a similar pattern (Jones et al., 2008; Yu, 2022).
Method
This research mostly followed what Yu (2022) did by analyzing the Science Productivity Higher Education and Research Economy Project (SPHERE) data. This contains publication information of the Web of Science’s Science Citation Index Expanded (SCIE) data (Baker et al., 2015). It also provides impact factors of journals from 1994.
This study analyzed research collaborations in 1980, 1990, 2000, and 2011. The analysis started from the 1980 dataset because research publication steeply increased from 1980 (Powell et al., 2017).
Impact factors were adopted to measure the quality of papers and journals (Glänzel & de Lange, 2002; Kim et al., 2010). This paper included the top 25% of journals in terms of impact factors. For 1980 and 1990, the 1994 impact factor was applied (Shamekhi et al., 2019).
This study conducted social network analysis which consists of actors and ties. Actors of this study were U.S. universities and foreign countries. In total, there were 906 U.S. universities and 187 countries in the analysis. While U.S. universities were categorized by research university and TNRs by applying the 1994 Carnegie Classification, which is the earliest version accessible, for all years to make the categories constant (Shamekhi et al., 2019), this research did not distinguish types of organizations that were not in the U.S. Secondly, when a U.S. university published a research paper with any institution in a foreign country, a tie was created.
Matrices analyzed were one-mode symmetrical, undirected, and dichotomous. If one U.S. university published at least one article with a foreign country together, 1 was assigned to both of them. If a U.S. university published no papers with international organizations, 0 was assigned. Therefore, degree centrality, in this paper, indicates the number of U.S. universities that a country collaborated with or the number of countries that U.S. universities worked with.
This study examined density, subgroups of the networks, degree centrality, betweenness centrality, eigenvector centrality, the Gini coefficient, and QAP regression. For subgroup analysis, because matrices were large, this study chose the Clauset-Newman-Moore method (Clauset et al., 2004). Although graphs cannot be included, the networks were visualized with the Horal-Koren algorithm by using NodeXL.
This study calculated the Gini coefficients, which are widely used to measure inequality (Kim et al., 2021), of degree centrality of U.S. universities to study whether international ties formation became more lopsided over time (Kim & Kim, 2017).
Finally, the QAP regression studied whether a dyadic independent variable predicts a dyadic dependent variable (Borgatti et al., 2018). The independent and dependent variables were the valued 1980 network and the valued 2011 network respectively (Yu, 2022).
Result
From 1980 to 2011 in the top 25% journals, the number of U.S. universities in the network increased from 212 to 813. The number of countries increased from 70 to 172. On average, the U.S. universities worked with six countries in 1980 and 19 countries in 2011. On average, foreign countries collaborated with 18 U.S. universities in 1980 and 88 U.S. universities in 2011.
The number of subgroups decreased from eight to six. In all four years, density was about 0.03. The modularity scores remained below 0.30. The subgroups were overlapping.
Harvard University (hereafter Harvard) had the highest degree centrality in all four years. Harvard worked with 29 countries in 1980 and 112 countries in 2011. Canada and the U.K. were major collaborators of U.S. universities.
Harvard, Canada, and the U.K. had the highest betweenness and eigenvector centralities in all four years. China became a significant actor in 2011. They were on the shortest paths between other nodes and were connected to other nodes that had a large number of ties (Borgatti et al., 2018).
This research compared the average degree centrality of research and traditionally non-research universities (TNRs) in the U.S. The number of TNRs rose from 93 to 689 during the three decades. However, research universities’ degree centralities more dramatically increased from eight to 62.
The Gini coefficients of U.S. universities’ degree centrality increased over time from 0.49 in 1980 and 0.62 in 2011. As scientific knowledge production increased, international collaboration was unequally distributed.
The coefficient of the valued 1980 matrix was positive and statistically significant in the QAP regression. Actors that had a tie in 1980 were more likely to have a tie in 2011. Degree centralities, the Gini coefficients, and the QAP regression result suggest that research universities dominated international collaborations. The Matthew effect (Merton, 1968) existed in the networks.
Conclusion
This study had similar results from what Yu (2022) found. The number of U.S. universities in international collaborations steeply increased over time in the top journals. The number of countries in collaboration increased. Various institutions participated in international research production. However, research universities overwhelmed the international collaborations in the top journals. TNRs in the U.S. had smaller international ties when they published papers in prestigious journals. Research universities in the U.S. became more dominant in knowledge production. International collaboration system was disadvantageous to TNRs.

Author

• Wan Yu, The Pennsylvania State University