Countries analysed

28

Study period

2004 – 2024

Threat typologies

4 clusters

Security regimes

11 periods
About this application

Research question: Do EU member states adjust their parliamentary defence rhetoric in response to external security threats, and do they differ in how responsive they are?

This application presents the interactive replication of:

Penchev, G. (2026). Political Debate as a Filter: Defence Policy Dynamics in EU Member States, 2004–2024. Penchev, G. (2026). Political debate as a filter: defence policy dynamics in EU member states (2004-2024). Eastern Journal of European Studies, 17(01), 216-243. https://doi.org/10.47743/ejes-2026-0109

Method: The monthly Regional Threat Index is built from UCDP GED conflict fatality data (1992–2024), detect structural breaks using the PELT algorithm, and classify 11 security regimes. We extract annual government and opposition defence stance scores from Manifesto Project party manifestos (variable per104 — Military: Positive) combined with ParlGov cabinet data. We then compute Dynamic Time Warping (DTW) distances between stance and threat series to measure how closely each country's political debate tracks external threat dynamics.

Clusters: NbClust majority-vote k-means clustering on the three DTW metrics identifies four threat-response typologies and two spending-alignment typologies.

Navigate using the tabs above:

  • Threat Index — UCDP conflict fatalities, security regimes, country-specific exposure
  • Debate & Spending — EU trends in defence rhetoric and actual spending
  • DTW Metrics — per-country alignment scores and robustness checks
  • Country Typologies — interactive map, cluster profiles, and alluvial flow

Data sources
Dataset Variable Source
UCDP GED v26.1 Conflict fatalities ucdp.uu.se
Manifesto MPDS2025a per104 manifesto-project.wzb.eu
ParlGov Cabinet composition Harvard Dataverse
Eurostat GDP, COFOG defence ec.europa.eu/eurostat
SIPRI via WDI Military spending data.worldbank.org
Replication

All scripts and instructions are available at:

gpenchev/art

Pipeline: R · tidyverse · changepoint · dtw · NbClust

UCDP GED data © Uppsala University (CC BY 4.0). Manifesto data © WZB Berlin. App: MIT Licence.

Y-axis scale:
Regional Threat Index — Europe's neighbourhood (1992–2024)
Security regimes
11 structural breaks detected by PELT algorithm (BIC penalty, min. segment = 24 months). Click a row to highlight on chart.
Country-specific threat exposure
Distance-weighted fatalities from conflicts in Europe's neighbourhood, weighted by inverse great-circle distance from the country's capital.
All series normalised to [0–1] for comparability. Manifesto Project per104 (Military: Positive) averaged across all EU-27 member states.
Stance = Manifesto per104 (% quasi-sentences on military positively). LOCF imputation applied between elections. Countries grouped by threat cluster type in selector.
DTW metric space
Stance vs Threat alignment
DTW metrics by country (Fig 5 replica)
Correlation with main analysis
Debate intensity is perfectly stable (r = 1.0) — it depends only on gov vs. opp stance, not on the threat index choice.
Silhouette score by k
Display
Click a country to see its profile
PCA of DTW metrics — cluster separation
Threat → Spending cluster flow (Fig 10)
Country typology summary