Beyond the Winner: Stage-Specific Candidate Choice Sets and Political Careers in National Executive Elections

This repository provides the data and code for Beyond the Winner: Stage-Specific Candidate Choice Sets and Political Careers in National Executive Elections. The study examines three distinct electoral outcomes - eventual executive victory, qualification for a subsequent ballot, and decisive-stage victory - and investigates whether electoral performance predicts a defeated candidate’s return at the next election. It distinguishes the information available before an election from performance observed after the ballot. The collection covers 140 electoral cycles and 202 held stages in 32 countries during 1995-2024. Each application uses a sample defined by its electoral rules, outcome requirements, and evidence coverage. The primary career analysis includes 442 defeated-candidate origins across 55 elections in 21 countries, with 122 subsequent returns. Candidate and party histories use only information preceding the relevant prediction date. Career outcomes distinguish an observed nonreturn from an unavailable electoral opportunity, incomplete candidate roster, or unresolved identity link. The analyses combine penalized choice and logistic models with country-held-out prediction, country-level resampling, and sensitivity analyses addressing institutional differences, source coverage, identity linkage, sample composition, and alternative successor-election definitions. Predictive associations are interpreted separately from causal effects, and uncertainty and unavailable comparisons remain explicit. The package supports two replication routes. The processed-data route reconstructs the analytical panels and reproduces 28 tables and seven figures. The raw-source route provides source locations, download and scraping tools, and detailed instructions for reconstructing candidate identities, party continuity, electoral opportunities, and outcome coding. Included materials comprise processed coding records, source evidence, variable definitions, estimator specifications, fixed random seeds, and numerical benchmarks. The processed-data analysis runs offline once the required Python dependencies are installed.