Methodology

Forecast of the Missouri General Assembly election of Tuesday November 3, 2026: all 163 House seats and the 17 even-numbered Senate districts. Written 2026-11-01. Judgment calls are numbered D-01 to D-10 in DECISIONS.md.

What is forecast

For each seat, the probability each party wins. For each chamber, the full distribution of seats, and the probability Republicans keep a veto-proof two-thirds: 109 of 163 in the House, 23 of 34 in the Senate (the 17 Senate seats not up are fixed: 10 Republicans, 7 Democrats).

Why a fundamentals model

No district-level polling exists for Missouri legislative races (one Republican internal in SD 8 is the only poll found). State legislative results are, however, highly predictable from the district's partisan lean, the national environment, incumbency and whether the race is contested at all. Under the current map the 2024 district lean alone explains the 2024 House results with a residual of about 2 points of two-party share.

The seat model

For seat d, the expected Democratic share of the two-party vote is

E[y_d] = 0.5 + a_2024 + shift + b (L_d - 0.5) + c_inc + k r_d + e edu_d + q prim_d

• L_d: the district's 2024 lean, a weighted mean of its Democratic two-party share for governor (one half), president and US Senate (one quarter each) (D-01, D-12), computed from Secretary of State precinct results allocated to the 2022 district plans. The governor race predicts legislative voting best out of sample.
• a_2024, b, e: fit by least squares on contested House seats in 2022 and 2024 (year intercepts, one slope, an incumbency term and an education term). Fitted values: slope 1.00, education -0.043 share per unit of bachelor's share (D-19).
• c_inc: incumbency. The symmetric Missouri estimate is 1.3 to 1.6 share either way; in a wave year the wave party's incumbents gain more than the other side's lose (Virginia 2025: +2.9 vs -1.0 around open seats). Blended at half weight: Democratic incumbents +2.2, Republican incumbents 1.2 (D-05, D-17).
• q prim_d: the district's August 2026 primary Democratic vote ratio minus its lean, de-meaned statewide, at 0.03 share per unit (D-18).
• shift: the expected change in the Missouri legislative environment from 2024 to 2026, +3.3 share with a t(5) scale of 1.5 (D-03). It comes from the late-October generic ballot (about D+8.7) translated through Missouri's historical elasticity of about 0.58 to the national House swing.
• r_d: the seat's residual from the 2024 fit (or the 2022 residual decayed if the seat was uncontested in 2024), carried with k = 0.75 when both candidates are the same as last time, 0.65 when the incumbent returns, 0.40 when the previous incumbent is gone (D-04). Missouri's own 2022-to-2024 record gives slopes of 0.72, 0.36 and 0.43 for those cases. This is how a popular incumbent who runs ahead of the district lean keeps most of the credit, and an open seat keeps some.

Uncertainty and correlation

Each of 100,000 simulations draws: one statewide shift (t5, scale 1.5 share), one error per region (normal, 0.75 share; seven regions), one error per district type (normal, 0.75 share; urban core, inner suburban, suburban, exurban, rural; D-15), and one error per seat (t7). The seat scale starts from the out-of-sample backtest rmse (1.7 share) and rises with competitiveness: 1.3x at an even lean, 0.75x beyond 20 points (D-13), x1.25 for open seats, x1.5 for seats with no contested result under this map (D-06). A Democrat wins a seat when the drawn share exceeds 0.5. Uncontested seats are fixed (D-07).

One race outside the legislature informs these shared errors. Missouri's 2nd congressional district (Ann Wagner v Fred Wellman) is the only polled partisan race that sits on top of the seats that decide the two-thirds line. Its polls, adjusted for partisan sponsors, run about 2 points more Democratic than the congressional fundamentals expect. That surprise is treated as a noisy reading of the statewide, St. Louis regional and district-type errors above, and the forecast is updated by the usual rules for combining a noisy observation with a prior: about +0.4 share statewide, a little more in the St. Louis suburbs (D-C5). It is not backtested; there are no district polls from earlier cycles to test it on.

Two Senate-district polls (SD 8 and SD 30, Remington Research Group for Missouri Scout, early October) are read the same way, as noisy readings of their seat's whole error: the statewide, regional and district-type parts and the seat's own. They are adjusted for the Republican-leaning sponsor and aged exactly like the MO-2 polls, and one legislative poll is given an error of 4 points of share (more than a congressional poll's 3.3). The MO-2 surprise and both polls update the errors together in one step, so evidence about the statewide mood is not counted twice: together they add about +0.5 share statewide, and SD 8 and SD 30 move a further +0.5 and +0.1 for their own polls (D-25). Election night starts from the same statewide shift. Like the MO-2 observation, this is not backtested: no district polls from earlier cycles exist; it is scored after the results.

Backtest

Fit on 2022, forecast 2024, and the reverse, on contested House seats. With the intercept a forecaster would have had (previous cycle's intercept adjusted by the national swing), the model called 107 of 110 seats in 2024 and 65 of 69 in 2022, Brier scores 0.023 and 0.029, rmse 1.7 share, against 104/110 and 65/69 for "district lean alone" (Brier 0.055 and 0.058). Applied to the 2022 Senate races (the even districts, whose leans are geometric estimates, D-11) with House coefficients: 11 of 11 called, rmse 0.020. The caveat: 2022 and 2024 had almost the same national environment, so the backtest validates the district model, not the size of the 2026 shift. The shift is the forecast's main uncertainty and is treated as such.

What is not in the model

The published ratings of other forecasters (D-09); they are shown beside our numbers as comparisons and will be scored alongside them. Campaign finance is on file but is not a term in any seat's mean: it only widens the seat error outside the competitive set (D-27). District polls are in the model (D-25). The Auditor polls inform the State Auditor's candidate effect (D-26). The St. Louis University/YouGov House generic ballot is published as a check beside the model's own expectation and is not applied to any mean (D-26). Candidate quality is scored for the competitive races (see D-20).

Scoring

Election-night model projections

Beginning with the November 3, 2026 needle release, a model projection requires at least a 97.5% chance for one outcome, evaluated before display rounding, and qualifying returns. This replaces the earlier 95% threshold. The existing 5% coverage gate remains; district projections require that district's returns, with an exception for an explicitly complete non-tied count. A high pre-election probability alone does not produce a call.

These are Missouri Election model projections, not certifications. Later returns can remove a projection or change the favored outcome. The homepage lists current projections as of the displayed model run; it does not present that run time as the first call time. A delayed feed retains the last validated figures with their original timestamps. The 97.5% threshold is a modeling policy, not a validated guarantee of a 2.5% error rate.

The chamber needles show probabilities of ending or retaining the Republican supermajority. The combined needle uses correlated joint simulations for ending it in at least one chamber. These aggregate probabilities are not marked as formal calls by the homepage. Competitive district needles show up to six contested races per chamber, selected from the published forecast where both major parties have at least a 5% chance. The selection stays fixed while their probabilities update. A shared statewide model adjustment is labeled separately from direct district returns; missing live district estimates are never replaced by an unlabeled old forecast.

Post-election evaluation

After certification: Brier score and log loss over contested seats, seats called correctly, calibration by probability bin, whether each chamber's seat count fell inside the 10th-90th percentile range; compared against State Navigate (Nov 2), On Point Politics (Nov 1) and a lean-only baseline. The forecast is re-run every hour and a new version is published only when its rounded figures change. The rule scores the last forecast published before polls close at 7 pm CT on election day, and the history of every published version is kept (data/forecast/forecast_history.json).