Data-Driven Regime Type Classification

This paper produces a theoretically rooted, data-driven typology of regime types. It draws on a variety of indicators of disaggregated regime properties and employs latent class analysis to identify empirically relevant regime constellations. Reaching the limits of dichotomous and one-dimensional continuous measures of democracy, current research focuses on regime distinctions within democracies, hybrid regimes and autocracies. Much progress has been made in this area, but most approaches are heavily deductive: They make theoretically informed decisions to select indicators. This is prudent practice, but it does not tap the full potential of today's large number of democracy-related indices that provide disaggregated indicators (such as details on elections, civil liberties, the effective power to govern, transition properties, leader properties etc.). This paper proposes an inductive approach that incorporates the disaggregate information from all available indicators simultaneously. It identifies indicators that make substantial contributions to distinguishing countries from each other. On the basis of these empirically relevant indicators, regime type constellations are identified by latent class analysis. This procedure can uncover distinctive properties and groups that have been overlooked in purely deductive approaches. It provides guidance to how many real-world regime types exist (as opposed to theoretically derived grids with lots of white space). It also helps determine thresholds within the employed indicators (as opposed to more arbitrary a priori thresholds that are set despite a lack of strong theoretical guardrails). Finally, the method allows the estimation of probabilities of belonging to a particular regime type (as opposed to deterministic cross tabulations of a limited number of ordinal variables). The paper estimates sets of regime types for various time periods (depending on the temporal coverage of the subindicators) and traces the movements of countries between regime types over time. Replications of extant studies demonstrate the value added of the empirical typology.