Forecasting Runoff Elections: A Parsimonious Model Using First-round Results
We present a simple model of runoff elections forecasting based on the distribution of votes in the first round. Using a collection of 128 presidential elections from 44 countries, our model indicates a strong positive relationship between the top two candidates’ first-round vote share and the second-round outcome. We confirm these results with an analysis of 287 elections for executive office in Brazil and evaluate our parsimonious approach next to more sophisticated specifications, but find limited improvement from including additional predictors. We argue the model performs well because (1) first-round results incorporate candidate and campaign-specific characteristics, (2) abstention rates increase slightly across rounds, (3) trailing candidates often need a large share of the remaining vote, and (4) voters of first-round losing candidates rarely have homogenous-enough preferences to turn elections around. As the model requires only readily available information, it can be easily applied to different contexts.