Network Structure and Ideological Polarization in Policy Sciences: Evidence from Co-authorship Networks

• In Event: Methods Student Research Panel

Abstract

Introduction/Background While extensive research has documented how social media networks contribute to political polarization, the relationship between academic collaboration networks and scholars' ideological positions remains largely unexplored. This gap is particularly significant for understanding knowledge production in policy sciences, where ideological perspectives may shape research agendas and findings. Previous studies have been limited by incomplete network data and challenges in measuring ideological orientation at scale across policy research communities.

Purpose/Research Question This study examines whether and how co-authorship network structures influence policy scientists' ideological trajectories. We investigate three key questions: (1) Do closely-knit academic communities exhibit greater ideological homogeneity in policy research? (2) Does network centrality correlate with ideological extremity among policy scholars? (3) Can interdisciplinary collaboration moderate polarization effects in policy sciences?

Methods We conducted a comprehensive content analysis of the Microsoft Academic Graph (MAG) by December 2021, comprising 4132651 of policy sciences papers authored by researchers connected by 7028641 co-authorship edges. We constructed an undirected network with edge weights representing collaboration frequency and applied the Louvain algorithm for community detection, identifying numerous distinct communities with a high modularity score indicating highly segregated collaboration clusters. To measure ideological orientation, we employed large language models (LLMs) to classify paper titles on a liberal-conservative spectrum. Our analytical framework incorporated both author-level metrics (degree centrality, clustering coefficient, k-core value, neighbor degree, local homogeneity) and community-level features (intra-edge ratio, density, maximum k-core). We analyzed relationships between network position and ideological stance using multivariate regression models while controlling for academic impact, productivity, gender, and disciplinary breadth.

Results/Findings Our analysis reveals a strong association between network clustering and ideological extremity in policy sciences. Higher clustering coefficients consistently correlate with more extreme ideological positions across multiple specifications. This pattern suggests that researchers working within tightly connected collaboration groups tend to adopt more pronounced ideological stances in their scholarship. Importantly, we find that scholars publishing across more disciplines exhibit significantly lower ideological extremity, indicating that interdisciplinary engagement may moderate polarization effects. At the community level, higher intra-edge ratios and greater density correlate with stronger ideological leanings in either direction, while communities with deeper maximum k-cores tend to demonstrate more moderate positions overall.

Conclusion/Implications This study demonstrates that dense clustering in academic co-authorship networks amplifies ideological extremity, mirroring echo chamber effects observed in social media but within formal academic collaboration structures. Conversely, interdisciplinary scholarship consistently correlates with more moderate political attitudes. These findings suggest that promoting cross-disciplinary collaboration may not only foster innovation but also enhance ideological diversity within policy science research. For funding agencies and academic institutions, these insights highlight the importance of supporting interdisciplinary initiatives as a mechanism for reducing polarization in policy-relevant scholarship and improving evidence-based policymaking.

Author

• Wenjing huang, The Hong Kong University of Science and Technology (Guangzhou)
  Presenting Author