Replication Data for: Citizen forecasts of Mexican presidential elections, 2000--2024

Scientific election forecasting has a long and growing tradition in established democracies, but has not made much progress in Latin America. To make progress, we examine citizen forecasts of Mexican presidential elections from 2000 to 2024. We develop novel statistical tools to study whether citizen forecasts are better at forecasting than chance and than vote intention polls. Using 55 surveys we find that citizens became better over time. While they forecasted who wins worse than chance at the dawn of democracy in 2000, they forecasted most recent presidential elections correctly and are now on par with vote intention polls. While citizens are also better than chance at forecasting vote shares, they remain, however, worse than vote intention polls. To understand why, we analyse the bias and variance of citizen forecasts for the first time. We find that the main reasons for their worse performance are socialisation under a one-party rule, inexperience with free-and-fair elections, and unwarranted concerns about the cleanliness of the elections.