About the Model — Methodology

Election Probabilities is a machine-learning forecast for 2026 U.S. primaries, House, Senate, and governor contests, built to find an edge in down-ballot races where prediction-market attention is thin and polling is sparse.

How it works

Public data — fundraising, endorsements, polling, and historical results — is collected and cleaned into one table. Each candidate becomes a row of features (cash on hand, share of money raised, weighted endorsements, district lean, incumbency). A model trained on past primaries and general elections scores every candidate, and the raw scores are calibrated into honest win probabilities that sum to 100% per race. Predictions refresh daily.

What a win probability means

A win probability is each candidate's estimated likelihood of winning their specific race — not their projected vote share. A candidate can lead in raw votes in a crowded field and still have a lower chance of finishing first. Within each race, probabilities sum to 100%.

Why down-ballot

The focus is on races where prediction-market attention is thin — down-ballot and primary contests where the crowd is less likely to have priced things correctly, and where a model has the best chance of an edge.

Common questions

Is a win probability the same as a vote-share prediction?

No. A win probability is the chance a candidate finishes first in their race. Two candidates can be tied on projected votes but have different win odds depending on how the rest of the field splits.

How often is the forecast updated?

Predictions refresh daily, and the underlying model is retrained as new fundraising, endorsement, and polling data become available — most heavily in the weeks before each election.

Why focus on primaries and down-ballot races?

Presidential and top-line races are heavily covered and efficiently priced. A model adds the most value where attention is thin — primaries, House districts, and governor races the crowd ignores.

What happens after a primary is decided?

The general-election forecast is seeded from predicted nominees, then updated with the actual winner once the primary is called.

What data does the model use?

Fundraising (FEC and state campaign-finance portals), endorsements, available polling, district and state partisan lean, incumbency, and historical results from past primaries and general elections.

See the full accuracy report · 2026 race forecasts

Last updated July 23, 2026.