Several weeks prior to the end of the 2022 election cycle, Voter Science produced a set of general election predictions for legislative district candidates. Using a combination of demographic factors and expected voter turnout, we produced these predictions at the individual candidate level using regression based statistical modeling techniques. These predictions were shared with a small circle of trusted stakeholders prior to the general election, and we delayed releasing them to a larger audience to avoid an undue influence on legislative contests.
On the afternoon of the general election, we released our predictions to the general public. Our original predictions can be accessed by clicking here: 2022 Legislative Race Predictions
Post election, we compared our predictions with candidate level performance. On average, our predictions tended to be highly accurate: the mean prediction error (or MAE) was only 2.5%. Of the 94 races assessed, we were able accurately predict the outcomes of 97% of the time.
The final comparison between our predictions and candidate performance can be found here: 2022 Comparison of Voter Science General Predictions and Actuals.
Our ability to accurately predict races ahead of time has several applications including resource allocation across multiple races as well as assessing the viability of different candidates to run for office. We will cover potential applications of this ability in a future blog post.