Mapping the Complexity of Democracy: Entropy-Driven Regime Discovery with Machine Learning and Explainable AI

Democracy is a multifaceted institutional phenomenon that cannot be comprehensively captured by a single index. This paper defines unique institutional configurations in various countries according to six V-Dem democracy dimensions, including the Liberal Democracy Index, Electoral Democracy Index, Liberal Component Index, Egalitarian Component Index, Participatory Component Index, and Deliberative Component Index. It uses data on 179 countries and conducts a primary K-means clustering method, supplemented with Hopkins statistics and gap statistics, bootstrap stability tests, and Gaussian Mixture Model (GMM) robustness tests. The paper also uses Shannon entropy with Miller–Madow correction to measure a regime’s institutional heterogeneity, and the Rule of Law Index of the World Justice Project for external validation. As a result, the paper uncovers seven empirically identifiable democracy regimes featuring different institutional configurations. Different regimes have varying levels of internal institutional heterogeneity, being higher in hybrid regimes, lower in consolidated regimes. WJP analysis reveals additional systematic differences in rule of law outcomes in different regimes. Overall, the results are in line with the multidimensional approach to democracy and show how unsupervised learning and information-theoretic analysis may complement democracy indices.