Demography, Early Voting, and Election Integrity in South Korea: Evidence Across Four Electoral Cycles

Allegations of fraud in South Korean elections have centered primarily on the in-precinct early-voting channel, which consistently produces higher democratic Party vote shares than same-day ballots. We assess these claims using a unified framework applied to four elections spanning 2020-2025: the 21st General Election (2020), the 22nd General Election (2024), the 20th Presidential Election (2022), and the 21st Presidential Election (2025).First, we apply several forensic statistical tests - including simulation-adjusted second-digit Benford's law, last-digit uniformity, mixture-model fingerprinting, and evaluation of the widely-cited 63:37 early-vote ratio claim - and find no systematic pattern indicative of large-scale vote manipulation, while flagging a candidate-specific last-digit anomaly in the 2022 presidential election that warrants ongoing scrutiny.Second, we regress Democratic vote share and early-vote share on principal components derived from census, employment, and housing data at the sub-district (dong) level, with province-interacted election fixed effects and population-weighted least squares.Across all elections, sociodemographic characteristics explain 73-94% of the geographic variation in vote shares, and the same predictors perform equally well for early votes and same-day votes.For the 2025 presidential election - a three-candidate contest - we instrument the third-party candidate's vote share using the 2017 conservative-reform candidate's support, recovering a causal estimate of vote diversion.Our findings suggest that geographic patterns in early voting are well-explained by who votes early, not by when ballots are counted.