IFSK Department Analysis
Introduction
The aim of this report is to provide an overview of political science research at a range of European universities, specifically the topics which are covered in their research and the topical profiles of the individual universities. The topics are inferred algorithmically from the abstracts of published journal articles, in an effort to represent the research as closely to the produced research as possible, without imposing assumed epistemic structures on the text corpus.
Topic profiles allow us to gain an understanding of how the research portfolios of universities vary, which groups of universities overlap, and where experts can be found on topics that are not our own focus.
Methodology
We use the journal categories Political Science, Public Administration, and International Relations from the Web of Science as basis for the analysis. All journals from the former categories are included, but a selection of political science specific journals has been handpicked from International Relations. The analysis includes all journal articles and reviews from 2012 to 2022 (n = 35,459). However, not all these articles have abstracts registered in the Web of Science, and only 26,651 (75.2%) publications are included in the final analysis.
All titles and abstracts are concatenated and transformed with SciBERT (Beltagy, Lo, and Cohan 2019), a natural language model pretrained on scientific documents, to embed words of a document on 770 dimensions. This gives us better data for grouping documents based on their conceptual content rather than the literal textual representation of the content. We then tested both UMAP (McInnes, Healy, and Melville 2020) and t-SNE (Maaten and Hinton 2008) for dimensionality reduction and visualization of the results, and found t-SNE to provide clearer global and local structures in this case. However, the resulting mapping of the area is denser than many other topic mappings, and the identifiable cluster numbers were highly sensitive to parameter settings. With some experimentation, we set the k nearest neighbours for the Louvain community detection algorithm to 200, in order to not have too small and specific topics, but also not too broad. This gives us 22 distinct clusters, with relatively little overlap, while earlier experiments with approximately 40 clusters had significant conceptual overlaps.
For the final step of the analysis, we use author keywords from the articles in each cluster, to describe the content of the clusters in a human readable format. We extract all keywords and count their occurrence across all documents (df) and within each cluster (tf), which we then use to calculate their cluster specific weight, tfidf, as tf/df, but excluding those keywords which occur very infrequently (1-3 times across all articles in a cluster of more than 1,000 publications). We give the 30 keywords with highest tfidf scores to ChatGPT, using GPT 4.0, to generate labels and short descriptions for each cluster.
Topic profiles are reported as frequency tables of articles by topic clusters; one university may have a strong profile in governance research, while another focuses on European union policies. All profiles use ratios of that university’s research rather than absolute numbers, to improve comparability between universities, even when the size of political science departments differ.
Topic Overview
ChatGPT Labels
Topic 1: Feminist and Identity Studies
Focuses on feminist theories, identity construction, everyday politics, social reproduction, and critical security studies.
Topic 2: Political Theory and Colonialism
Examines political theology, settler colonialism, counter-hegemony, decolonization, and Marxism.
Topic 3: Historical and Memory Studies
Explores the politics of memory, Enlightenment, World War II, Holocaust, and historiography.
Topic 4: Economic Policy and Trade
Analyzes firm-level data, international trade, fiscal consolidation, wages, and economic freedom.
Topic 5: EU Governance and Policies
Studies European administrative space, higher education governance, food safety, and EU external action.
Topic 7: International Relations and Conflicts
Focuses on rebel governance, Russian foreign policy, ASEAN, Iraq, and counterinsurgency.
Topic 8: Constitutional and Political Systems
Examines constitutional conflicts, indirect governance, growth models, fiscal federalism, and democratic backsliding.
Topic 9: Psychological and Behavioral Studies
Studies political orientation, social dominance orientation, intergroup contact, conspiracy beliefs, and evolutionary psychology.
Topic 10: Governance and Networks
Explores telecommunications, central government, inter-municipal cooperation, strategic planning, and regulatory agencies.
Topic 11: Research Methods and Impact
Analyzes Bayesian methods, reproducibility, citation impact, bibliometrics, and research performance.
Topic 12: Leadership and Public Management
Focuses on burnout, work engagement, transformational leadership, public service motivation, and HRM.
Topic 13: Ethics and Political Philosophy
Studies ethics of war, public justification, consequentialism, international justice, and relational equality.
Topic 14: Human Rights and Law
Explores the UN Convention on the Rights of the Child, migration law, European citizenship, and the European Court of Human Rights.
Topic 15: Climate and Marine Policies
Analyzes mitigation scenarios, marine planning, climate change policies, adaptation policy, and carbon pricing.
Topic 16: Elections and Voting Behavior
Focuses on correct voting, party affiliation, electoral expectations, female representation, and strategic voting.
Topic 17: Interdisciplinary Research and Governance
Studies interdisciplinary research, Earth system law, bioeconomy, global health security, and AI.
Topic 19: Economic and Financial Integration
Examines economic crises, banking union, sovereign debt crisis, and European Central Bank policies.
Topic 21: Military and Strategic Studies
Focuses on nuclear deterrence, counter-insurgency, proxy wars, cyberwar, and NATO strategies.
Topic 22: Legislative Processes and EU Politics
Analyzes EU directives, parliamentary voting, legislative organization, party cohesion, and parliamentary questions.
Results
University comparison
We use the university loadings on topics for comparing how similar the profiles of each university are. As opposed to the previous graph, where there is little overlap between topics, every university has a loading on each topic, and as a result the similarities are on average much higher. In the graph below, we only remove edges with similarity < .8. The majority of universities are centered around a cohesive core, with quite similar profiles. In the north of the graph, the majority of British universities are found, with apparently distinctly different profiles from the core. An important exception here is the University of Oxford, which is placed in the core. The Department of Political Science at Aarhus University is labelled IFSK, while all research from Aarhus University in the study population is labelled Aarhus Univ. This is mainly research from IFSK with some additions from Economics, Management and similar social science departments. IFSK is placed in the opposite end of the graph, compared to the British universities.
The graph below shows the full range of university topic similarities for IFSK.
Individual university profiles
Cent European Univ
ETH Zurich
Freie Univ Berlin
Ghent Univ
Goethe Univ Frankfurt
Heidelberg Univ
Humboldt Univ Berlin
IFSK
Katholieke Univ Leuven
King’s Coll London
Leiden Univ
London Sch Econ & Polit Sci
Ludwig-Maximilians Univ München
Lund Univ
Pompeu Fabra Univ
Radboud Univ
Sci Po
Stockholm Univ
Trinity Coll Dublin
Univ Amsterdam
Univ Antwerp
Univ Barcelona
Univ Bergen
Univ Bern
Univ Cambridge
Univ Coll London
Univ Copenhagen
Univ Edinburgh
Univ Essex
Univ Exeter
Univ Geneva
Univ Glasgow
Univ Gothenburg
Univ Hamburg
Univ Helsinki
Univ Konstanz
Univ Münster
Univ Oslo
Univ Oxford
Univ St Gallen
Univ Strathclyde
Univ Sussex
Univ Vienna
Univ Warwick
Univ Zurich
Uppsala Univ
Utrecht Univ
Vrije Univ Amsterdam
Vrije Univ Brussel
Impact
Mean Normalized Citation Score
Proportion of papers in top 10%
Number of papers in top 10%
Specialization degree
The figure below shows to which degree the research at a given university is focused on few topics, or balanced between topics. This is measured using the Herfindahl-Hirschman index, \(HHI\):
\(HHI = \sum_i^n p(i)^2\)
, where \(p(i)\) is the proportion of a university’s research in a given topic. The index ranges from complete balance at \(1/n\) (in this case \(1/18\), indicated by the green line) to 1, in which case all publications are on only one topic - complete imbalance.
References
Beltagy, Iz, Kyle Lo, and Arman Cohan. 2019. “SciBERT: A Pretrained Language Model for Scientific Text.” https://arxiv.org/abs/1903.10676.
Maaten, Laurens van der, and Geoffrey Hinton. 2008. “Visualizing Data Using t-SNE.” Journal of Machine Learning Research 9 (86): 2579–2605. http://jmlr.org/papers/v9/vandermaaten08a.html.
McInnes, Leland, John Healy, and James Melville. 2020. “UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction.” https://arxiv.org/abs/1802.03426.