Machine Learning Seminar: Votes, points, and point estimates: the central role of uncertainty in election forecasting
Reception: 3:00 p.m.
Seminar: 3:30 p.m.
When predicting election outcomes, it is natural to focus on increasing precision of forecasts. This precision can be gained via more survey data collection, more expensive and varied survey modes, and more sophisticated modeling techniques. Uncertainty of these predictions, while critical, can be amorphous, complex, and unknowable. However, the interrelated nature of elections, as played out from the electoral college to local boards, makes election forecasting a question of estimate uncertainty as much as estimate precision. This talk outlines the implications of different sources of uncertainty on election forecasting and possible methods to account for the types of cross-election bias that could affect predictions and resource allocation.