← Back to Dashboards
📖 About This Project
"Does the European Court of Human Rights Treat Countries Differently?"
Answer: YES - Systematic Country Differences Exist

A comprehensive statistical analysis of 1,904 substantive ECtHR cases (1968-2020) examining systematic country differences in violation findings using multi-method triangulation approach.

🌍 Regional Gap
Eastern Europe: 93.9% violation rate
Western Europe: 72.2% violation rate
Difference: +21.6 percentage points (p < 0.001)
📊 Country Effects Persist
9/16 countries (56.2%) remain statistically significant after controlling for article type, year, and applicant type. Odds ratios range from 3.49 (Croatia) to 32.45 (Ukraine).
⚖️ Judge-Independent
Country effects persist even after controlling for individual judge identity. 140 judges show average +29.1 pp higher violation rate for Eastern Europe (t=14.07, p<0.0001).
🤖 ML Temporal Validation
Random Forest trained on 1968-2014 data, tested on 2015-2020:
89.4% accuracy, AUC-ROC = 0.813
Indicates stable, predictable patterns across time periods.
🎯 Cross-Validation Performance
Random Forest best overall performer:
AUC-ROC = 0.808 (5-fold CV)
Demonstrates robust predictive patterns across different data splits and configurations.
📈 High Overall Violation Rate
Overall violation rate: 89.1%
Indicates strong case selection effects—only cases with merit survive domestic remedies and reach final ECtHR judgment.
🔬 Multi-Method Triangulation Approach
  • Exploratory Data Analysis: Descriptive statistics, temporal trends (1968-2020), and violation rates by country/region/article type across 1,904 substantive cases from 45 countries
  • Hypothesis Testing: Chi-square tests, two-proportion z-tests, and Cohen's h effect size analysis comparing Eastern vs Western Europe (p < 0.001, large effect size)
  • Logistic Regression: Country-level models (≥30 cases/country) with L1 regularization (Lasso) controlling for article type, judgment year, and applicant type (Pseudo R² = 0.226, AUC = 0.801)
  • Judge Fixed Effects: Advanced models testing whether country effects persist after controlling for 403 unique judges (171 judges with ≥20 cases analyzed in detail)
  • Machine Learning: Random Forest, XGBoost, and Gradient Boosting (≥30 cases/country) with 5-fold cross-validation and temporal validation (train: 1968-2014, test: 2015-2020)
⚠️ Important Caveats & Limitations
Statistical significance ≠ Discrimination. These findings do NOT necessarily indicate judicial bias or discriminatory treatment by the Court. Systematic differences may reflect legitimate factors including:
  • Genuine human rights conditions: Rule-of-law variations and actual human rights situations differ across countries
  • Unmeasured case characteristics: Case complexity, quality of legal representation, and strength of evidence not captured in available data
  • Selection bias: Only cases surviving domestic remedies and passing admissibility thresholds reach ECtHR final judgment stage
  • Post-communist transition challenges: Observational data cannot distinguish between structural challenges vs discriminatory scrutiny
This analysis establishes correlation, not causation. We demonstrate systematic patterns exist, but cannot determine their underlying cause without experimental manipulation. See full documentation (REFLECTION.md) for detailed methodological limitations.
📄 Read Full Documentation on GitHub