About
Machine Learning Winds builds probabilistic election forecasts, starting with the United States and expanding to elections that get little data coverage in English.
Forecasts here are probabilities, not predictions. A candidate with a 70% chance should lose about three times in ten, and a forecast that never misses is a forecast that is too timid.
We publish our methods and test them against past elections before trusting them with a live race. When a forecast is a backtest built after the fact, we say so at the top of the page.
Machine Learning Winds LLC is based in Colorado.