Methodology · US House
US House forecast methodologyBeta
Follows the global methodology, with the deltas below. Source spec: MKT-USM v3.5 (locked, beta contract).
What this is
A 2026 US House forecast for all 435 districts. It is published as a beta structural baseline, not a locked production forecast. Most districts (~386 of 435) are structurally modelled; it is lightly poll-adjusted only where district polling exists. Read the control probability with wide uncertainty.
How it differs from the global standard
- •Structural prior per district: a partisan baseline (PVI) on the enacted 2026 maps for all ten redrawn states (CA, FL, LA, MO, NC, OH, TX, UT, AL, TN), plus the national environment and incumbency. Where a district is polled, the polls are blended in (Bayesian update); where it is not, the structural prior carries it.
- •Monte Carlo: 40,000 simulations per chamber with a shared national-environment shock (widened to reflect historical polling error as uncertainty, never a fixed partisan correction) and region-correlated shocks.
- •Interim map input: Tennessee currently uses President-2024 as a stop-gap pending the full composite; the other nine states use a 2024-ending measure. This is disclosed and will be refreshed.
- •Candidate quality is dormant — no candidate-quality score is currently applied; it activates once nominees are clearer.
How to read it (beta caveats)
- •It is a beta structural baseline, not a locked production forecast.
- •Most districts are structurally modelled, not directly poll-backed.
- •It is lightly poll-adjusted only where district polling exists.
- •Maps are current to the latest verified 2026 enacted maps; some presidential-by-district inputs remain interim (Tennessee).
- •The House control probability carries wide uncertainty.
- •Candidate-quality modelling is limited and dormant until nominees are clearer.
- •The model will be recalibrated as more district polling and post-election validation arrive.
Validation
Held-out backtesting of the structural baseline is well calibrated; the model has a documented pre-election score ceiling because the production prior cannot be fully validated until there are real election results to test against. Current figures are on the live House tracker.