Every term, in plain words.

The range the seat count lands in for 8 of every 10 simulations. The remaining 20% of runs fall outside it, 10% on each side.

House: 229–262 Democratic seats means fewer than 229 or more than 262 each happen about 1 time in 10.

The starting margin MP-26 gives every race before simulating it: how the district or state voted for president in 2024 against the country, plus the incumbent's edge (in the House, part of how far they ran ahead of the 2024 presidential result) and today's national environment, blended with the race's own polls. The weights are fitted on the 2014–2022 midterms.

The model's chance that the race finishes very close, within a few points either way. Close races are the ones most likely to be called late.

The part of a forecast miss that hits every race in the same direction. When polls underestimate one party, they usually do it everywhere at once. In the model this shared miss is about ±2.9 points on each race margin.

It is why races are not independent coin flips: a bad night for Democrats in Ohio usually means a bad night in Maine too.

A national poll question: "If the election for Congress were held today, would you vote for the Democratic or the Republican candidate in your district?" It measures the overall political mood, not any single race.

D+4.4 means Democrats lead the average of these polls by 4.4 points.

Control of the House of Representatives needs 218 of 435 seats. Every seat is elected every two years.

The person, or party, currently holding the seat. An open seat is one where the incumbent is not running.

Who a poll interviewed. Likely voters are screened for how probable it is that they vote; registered voters are on the rolls; adults is everyone. Likely-voter polls get the most weight because they are closest to the real electorate.

Weights: LV 1.00 · RV 0.86 · adults 0.72.

The part of a forecast miss that belongs to one race only: a strong candidate, a scandal, a local issue. It is large for a single race (about ±9.6 points) but it cancels out across hundreds of races.

How far one party leads the other, in percentage points. D+4.4 means Democrats ahead by 4.4 points; R+2.0 means Republicans ahead by 2.0.

49.3% D vs 44.9% R is a margin of D+4.4.

The middle outcome of the 50,000 simulations: half of the runs give a party more seats than this, half give it fewer. It is the model's single best guess for the seat count.

The U.S. election held halfway through a presidential term. All 435 House seats and about a third of the Senate are on the ballot. The 2026 midterms are on Tuesday, November 3, 2026.

How much the national poll average has changed since the starting margins were set. MP-26 builds today's national environment into every race's starting margin, so there is no separate movement to add: it is zero.

MP-26, the forecast model behind this site. Every race starts from its own baseline (2024 presidential lean, incumbency and the national environment), blended with the race's polls, and the election is simulated 50,000 times. The current version is v0.5; it is experimental, and its coefficients are fitted on the 2014–2022 midterms.

How much a race follows the national mood rather than local factors. In MP-26 every race's starting margin moves point for point with the national environment, and 70% of a national polling miss is passed on to each race.

How much one poll counts in the average. Newer polls count more (weight halves every 14 days), bigger samples count more, and likely-voter polls count more than polls of all adults.

A market where people bet on outcomes, such as Polymarket. A price of 62¢ on "Democrats win" reads as a 62% chance. The site shows markets next to the model for comparison; they never feed into it.

A plain-word summary of how competitive a race is. Safe: not competitive. Likely: clear favourite. Lean: favourite, but an upset is realistic. Toss-up: either side can win.

Senate seats that are not on the ballot this year, because senators serve six-year terms. In 2026, 34 Democratic and 31 Republican seats are not up, so they are fixed in every simulation.

Democrats need 51 of 100 seats. At 50–50 the Vice President, a Republican, breaks ties, so Republicans keep control.

Playing the election many times with random but realistic errors, then counting how often each outcome happens. The model runs 50,000 elections per chamber; a 99% chance means one party won in 99 of every 100 runs.

An election to fill a seat that became vacant before the end of its term, for example after a resignation or death. It can put a state's second Senate seat on the ballot.

A hypothetical shift in the national vote applied to every race at once. The swing slider on the home page lets you ask "what if the country moved 3 points toward Republicans?"

Line up all 435 districts from most Democratic to most Republican. The 218th is the tipping point: whichever party wins it also wins every seat on its side of the line, and with them the majority.

If the tipping point is D+7.7, Republicans need to win districts where they trail by 7.7 points, which only happens if the whole country moves toward them at once.

The projected vote share of the Democratic and the Republican candidate. Usually they add up to nearly 100; where a strong independent or third-party candidate is running, they add up to much less.

How often a candidate wins across the 50,000 simulations. 78% does not mean a landslide: it means the candidate loses roughly 1 time in 5.