This report introduces a new forecasting tool that uses structural data in order to make predictions of the outcome of elections on a global scale, complementing most existing election forecasting tools which are country-specific.
Initial modelling efforts yield a ~70% accuracy on testing data, indicating that our model is able to produce useful estimates of the probability of incumbent victory on new data. We also find that the length of a government’s tenure, the duration of democracy in a country and infant mortality rate are the most important predictors for incumbent victory.
Looking at directional relationships between our predictors and the probability of incumbent victory we also find that incumbents in very new democracies have a greater electoral advantage than incumbents in older democracies. We also find that government tenure can raise the probability of incumbent victory significantly but this advantage is not linear and reaches diminishing returns fairly quickly.