Our model depends primarily on polls to predict vote outcomes. Past polling error informs our modelling of the uncertainty of our predictions. We model the probabilities using the beta, weibull, and logistic distributions. The Electoral College vote and probability is estimated using 20,000 Monte Carlo simulations.
Polls prove predictive even early on in the presidential election process. 100 days before election day, presidential polls have a Mean Absolute Error of 5.3 points (Jennings and Wlezien 2016). This means that on average, polls of the presidential race accurately predict the final margin 100 days out within 5.3 points.
In addition to polls, which we weight as 2/3rd of our predictions in our forecast, we incorporate a unique model based on "Media Partisanship". It uses measurements gathered from Google Trends in order to asses media polarization and thus predict vote margins, and
it more predictive of past election outcomes than polls alone