US midterms 2026
Senate and House research, recorded polling, and prediction markets in one place.
The two chambers, together
The Stan research snapshot combines Senate polls and fundamentals with House district baselines. Shared election and state errors connect the races. Uncovered contests stay unresolved.
The research ranges assign uncovered contests to each extreme. They are coverage bounds, not confidence intervals or unconditional control odds. Reported same-party fields are conditional on the candidate lists being complete.
What the markets say
Separate provider prices, not model estimates. Source dates can differ from the research snapshot. See prices, history and settlement context →
What if both chambers move together?
Shift the Democratic-minus-Republican margin in every modelled race using the same recorded simulated elections. Same-party fields stay fixed; uncovered contests stay unresolved.
Negative shifts favour Republicans; positive shifts favour Democrats. This is a sensitivity overlay, not a new forecast. Recorded race estimates remain fixed.
Scenario data table
Model estimates and market prices are dated separately. Their differences describe these snapshots; they are not matched-time evidence of an edge. Blank estimates and explicit exclusions remain visible.
Candidate names are the reported field at the research cutoff. The market view retains provider candidate labels and observation times separately.
Prediction markets
Kalshi and Polymarket prices are shown independently. Contracts can have different settlement rules. Prices are not inputs to either research model.
Democratic Senate control
Democratic House control
Exact control-price observations
Balance of power
Senate / House control, shown for each platform. These independent contracts are not normalized into one distribution.
Democratic Senate seat buckets
Democratic House seat buckets
Bucket prices do not necessarily sum to 100%. Majority thresholds: 51 Democratic-aligned Senate seats; 218 House seats.
Senate race prices
| Contest | Provider candidates | Kalshi D | Polymarket D | Kalshi R | 30-day D change | Kalshi D history |
|---|
Governor race prices
Methods and recorded history
The broader Senate/House model is a dated research snapshot. The original Senate polling study keeps its own daily captures and scoring record. This shared interface preserves both studies and their source dates.
Two-chamber research: historical checks
Expanding-cycle fits hold out 2020, 2022 and 2024, each with at least four earlier cycles. These are selected D/R contests, not complete chamber-control backtests. Historical publication lags are reconstructed proxies. Three correlated elections do not establish calibrated control odds or an edge.
Original polling v1 on exactly matched historical Senate races
Earlier cycles only, under unchanged v1 code. The original pilot uses its own probability-clipping convention.
Full historical population and exclusions
Stan sampling diagnostics
Four chains; 1,000 warmup and retained draws per chain; adapt_delta 0.99. Publication requires zero divergences and depth hits, R-hat ≤ 1.01, bulk/tail ESS ≥ 400 and the energy check.
Original recorded Senate polling study
Latest eligible poll per firm within 45 days, equal firm weights, with shared election error and race-specific uncertainty. Missing, stale and multiparty races remain excluded. It does not provide full Senate-control odds. Historical validation did not establish better winner prediction than the fixed three-point polling baseline.
Original validation and uncertainty
Assumptions and sources
House fundamentals use earlier presidential votes re-tabulated onto the target election’s district maps, with national generic-ballot polling. The model estimates chamber-specific effects, correlated national cycle errors, shared state errors and separate race noise. It does not include House district polls, candidate or incumbency effects. Modelled races approximate D/R outcomes; minor-party wins are not simulated.
Current candidate lists are reported fields, not certified ballot captures. New independent winners are not assigned to a caucus. Senate arithmetic starts with 34 Democratic-aligned and 31 Republican holdovers. A 50–50 Senate is not a Democratic majority. Historical 2024 Senate polls use an Electoral-Vote supplement, a disclosed source change from 538.