Dynamic Bayesian forecast
Updated Loading…
Dynamic Bayesian forecast
Updated Loading…
Click any district for its posterior forecast.
National environment · House
These are distinct stages, not interchangeable versions of one number. Margins are Democratic minus Republican; positive values favor Democrats.
Economics inside the prior
| Metric | Model-oriented change | Weight | Contribution | Observation |
|---|
Each change compares a three-observation average with the same period one year earlier. Positive model-oriented changes favor the incumbent president's party; the economic coefficient translates the standardized composite onto the Democratic-margin scale.
Posterior learning
Economics sets a broad election-day prior. Silver Bulletin’s published likely-voter series is shown for context; the House likelihood is a separate, one-time weighted aggregate of Silver’s adjusted poll file.
Points show Silver Bulletin’s published likely-voter-adjusted average. The House update uses today’s separately labeled national likelihood once; historical averages are never stacked as independent evidence.
How the forecast works
The model starts with a broad economics-only expectation, updates it with one chamber-specific national polling input, and preserves uncertainty through every race and chamber simulation. Every margin is Democratic minus Republican: positive values favor Democrats and negative values favor Republicans.
Refresh Silver polling, five FRED series, and FEC/Clerk candidate records. Malformed or catastrophically stale data cannot replace the last valid forecast.
A standardized five-series economic index creates a deliberately broad pre-poll expectation. Its uncertainty includes the economic coefficient, input uncertainty, and a 3.5-point structural standard deviation.
The House collapses Silver’s adjusted poll universe into one weighted likelihood. The Senate uses Silver’s latest published likely-voter average once. The daily maintained-average history is never treated as independent repeated evidence.
Partisan lean, incumbency, national conditions, shared regional movement, and local error produce a probability distribution of margins for each race.
Only the latest verified Democratic-versus-Republican Silver average may update a race. Missing, third-party, or unresolved matchups remain fundamentals-only.
In 10,000 correlated simulations, national and regional waves move races together. Control probability is the share reaching 218 House seats or 51 Senate seats.
Do not confuse these numbers
Silver’s displayed likely-voter maintained average. It describes current sentiment and is context-only for the House.
One influence-weighted observation reconstructed from Silver’s adjusted poll file. This is what updates the House prior.
The poll-updated posterior after future-movement and election-error uncertainty is added.
-- means Democrats won the House in that share of the model’s 10,000 simulated elections. It is not a predicted vote share, a percentage of seats, or a guarantee.
Read the complete formulas, assumptions, validation, and glossary →
10,000 simulations preserve shared national and regional shocks.
Compare each race’s fundamentals prior with its Silver-average-updated posterior when available.
| Race | Incumbent | Prior | Posterior | Dem % | Average | Rating |
|---|
Daily probability of Democratic chamber control.