Every poll we use, and how we count it
Our forecast reads every public general-election poll of a 2026 Senate, governor or House race we can find. Before averaging them, it makes three corrections, all learned from how past polls compared with results.
2. The firm's track record. Some firms' polls have leaned toward one party over many elections. We shift each firm's polls by its past lean relative to other firms, shrunk toward zero for firms with few past polls.
3. Undecided voters. When many voters are undecided, the model allows for them breaking unevenly, based on how undecideds have broken in past races.
How we adjust each pollster
Every firm with a 2026 race poll. "Our correction" is how far we move each of its polls; it comes from the firm's record in past elections, so a firm without a record gets none. "Sponsored" counts polls paid for by a campaign or party, which are also shifted and half-weighted (correction 1 above). Generic-ballot polls are on the national page.
Every 2026 race poll
Newest first. The margin is between the two major candidates, with undecided and other respondents set aside (Osborn counts on the Democratic side in Nebraska). Click a race to see its polls charted against our average.
Poll data from VoteHub (CC BY 4.0), Decision Desk HQ and Wikipedia. Links go to each poll's release where the source provides one.
Poll fine print
Details from inside the polls, beyond the topline: independents, who voters expect to win, approval and more, as reported in The Turnout, our weekly newsletter. Newest first; each race's own page lists its details too.