Three US products, each with its own model and version.

The three US products

The House model has its own national-environment component, which does not replace the published generic-ballot estimate.

Generic ballot (us-full-v2.1)

The generic-ballot estimate is a weighted average of national polls asking which party people will vote for in the House election, recalculated each morning. A poll's weight is the product of five factors:

• Recency: exponential decay with a 21-day half-life, shortened to 10 days in the final 14 days before the election. Polls more than 365 days old are not used.
• Sample size: proportional to the square root of the sample size relative to 600 respondents, capped at 2.
• Pollster: 0.25 plus 0.75 times the pollster's reliability score, with a default score of 0.5 where none exists.
• Population: likely voters 1.0, registered voters 0.75, all adults 0.5.
• Partisan releases and tracking polls: each carries half weight.

The Democratic and Republican shares are the weighted means of the polls, and the margin is their difference. Each share carries an uncertainty that combines the spread of the polls, divided by the square root of the effective number of polls, with a fixed 1.5-point term that more polling does not remove, and a floor of 2 points. The published bounds are the estimate plus or minus 1.645 times that uncertainty. That would be a 90% interval if the errors were normal; it is not a posterior interval, and its coverage has not been measured.

Historical data note

Each US Midterms estimate was written on the morning it was produced. When a poll arrived later, the estimates for earlier dates were not recalculated, so many earlier daily estimates do not include every poll the model's own rules would count for that date. In a test of 1 to 28 August 2026, recalculated generic-ballot figures differed from those published by up to 0.38 points on the Democratic share.

From 3 to 15 September 2026 the generic-ballot estimate was recalculated each morning for the previous 28 days, updating rows from 7 August 2026 onwards. The historical House and Senate race estimates and chamber simulations were not recalculated; they are snapshots of what was published on each date, not a reconstructed backtest.

Some dates have no estimate, including every date from 16 September 2026 until the new models were released on 20 September 2026, when an automatic check stopped publication because the House estimate fell outside its expected range. From 20 September 2026 the House forecast uses us-house-hc-v3.1 and the Senate forecast SEN-A1-2026.1; estimates published before that date come from earlier models and are not directly comparable. The generic ballot continues on us-full-v2.1.

Interpreting probabilities

A 60% probability means that in the model's simulations the outcome happens in about 60 of every 100 runs. It describes the model's uncertainty, not a certainty. No probability is shown as 100% or 0%: above 99% it reads ">99%" and below 1% it reads "<1%".

Source attribution

Independent polls are sourced via the New York Times polling dataset, maintained by the NYT polling team. Aggregator running-averages from RealClearPolitics, Silver Bulletin, Decision Desk HQ, VoteHub, FiftyPlusOne and Race to the WH are listed for context and are not counted as independent polls. The running averages, model, methodology and seat projections shown on this site are FredCornell.com's own.

Version history