How this is built

The voters are the 32,730 respondents to the 2024 Cooperative Election Study, a 60,000-person academic survey released into the public domain, whose vote in 2024 was confirmed in state voter files. The study’s weights describe the nation, so each state’s voters are reweighted to the Census Bureau’s count of who voted there in 2024 by sex, race and Hispanic origin, and age. A statistical model gives each voter a chance of choosing the Democrat (or the leading independent where no Democrat ran) from their state, race, education, gender, age, income, type of area and religion, with extra terms for the South, so that a small group is not decided by a handful of answers.

Each state is then shifted as a whole, by the same amount for every voter in it, until the voters together reproduce that state’s actual 2026 result for the office. Every scenario on this page is recomputed from those voters. A House district moves by the same number of points as its state’s House vote; uncontested seats stay where they are. Senate totals include the seats not up this year.

Party identification is not used and is not offered as a group. In a single state’s sample the share of Democrats and Republicans within a group is often a sampling accident that no public count can correct, and using it moved some states in directions no other evidence supports.

What it cannot know: how each group moved between 2024 and 2026 differently from its state as a whole, and how the 2026 midterm electorate differed in make-up from the 2024 one. Both have to wait for the 2026 survey, expected in spring 2027, which will replace this one. A state is marked when the groups you pick leave fewer than 30 survey voters there, and every state carries a margin of error; when it is larger than the margin, the scenario cannot say who wins that state. Counts are provisional until states certify.

Sources: Schaffner, Shih, Ansolabehere and Pope, Cooperative Election Study Common Content, 2024 (Harvard Dataverse, CC0); U.S. Census Bureau, Voting and Registration in the Election of November 2024, tables 4b and 4c; 2026 counts as on our results page.