Work

Diffusion Maps and Sparse Reconstruction: A Two-Tier Approach to Ideal-Type Measurement

Year: 2026
Type: preprint
Abstract: Composite indices such as V-Dem’s Electoral Democracy Index (EDI) impose a priori weights on their component indicators; potentially obscuring how political regimes vary, particularly among non-democracies. We propose a data-driven alternative that uses regime similarity, rather than deductive aggregation, to characterize variation across the EDI’s disaggregated indicators. Applying Diffusion Maps, we derive a low-dimensional embedding without imposing a priori weighting. To render this embedding interpretable, we use LASSO regression to build a sparse reconstruction: the Spectrum of Regime Similarity (SRS) that recovers most of its structure from a handful of indicators, principally opposition autonomy and election quality. Validating the SRS against intrastate conflict onset and infant mortality, we find an inverted U-shaped relationship with both outcomes that the EDI obscures, and lower predictive error than the EDI for infant mortality, with comparable performance for conflict onset. Results demonstrate that regime similarity better discriminates among non-democracies and complements concept-driven democracy measurement.
Language: English
Open Access status: green
OpenAlex ID: W7225908039
Cites:
Cited by:
Related to: 10