Crosstab gathers every public poll of the 2026 elections into one place and runs them through the Crosstab model: a weighted polling average, pollster-lean corrections and a 20,000-simulation forecast. The same model code produces every number in the app, on the website, in the emails and in the daily briefing.
Crosstab aggregates publicly released polls from pollsters' own releases and public poll listings, including RealClearPolling and the polling tables on Wikipedia's 2026 election pages (CC BY-SA 4.0), which record each poll's field dates, sample size, voter type and margin of error. Every poll is matched against the ones already on file: the same poll listed by two sources, dated differently or under a different pollster name, is counted once. Each poll is shown with the pollster that conducted it. Polls sponsored by a campaign or party are included, identified where known, and weighted down.
For each race, Crosstab takes the most recent poll from each pollster over the past 45 days (up to eight pollsters, so no single firm can dominate by polling often) and weights each one by:
National averages (presidential approval, the generic congressional ballot) use the most recent poll from each pollster over the past 14 days. Trend charts show the average exactly as it stood on each day.
Every polling firm has a house style. For each pollster, Crosstab measures the average gap between its polls and other firms' polls of the same contests within 30 days (14 for approval), shrunk toward zero so firms with few polls can't earn large leans. A lean describes a firm's style relative to other polls, not its accuracy. The forecast subtracts each poll's lean before combining polls.
In each race, the lean-adjusted weighted average is blended with a starting estimate from published expert ratings, worth about 1.5 polls, so a single poll can't swing a race. The more polls agree, the more the average counts; the model uses the effective number of polls after weighting, so several stale polls don't masquerade as fresh ones. Uncertainty includes about 5 points of normal polling error on the margin, the disagreement between polls, and more the further we are from Election Day. Senate control comes from 20,000 simulated elections with a shared national error, so races can miss in the same direction together. Republicans keep control at 50 seats because the Vice President breaks ties.
Crosstab's ratings come straight from the forecast: Toss Up when either side's chance is 40–60%, Lean 60–80%, Likely 80–95%, Solid above 95%. Races without polling keep their starting rating.
Betting odds come from Polymarket's public prices (the midpoint of the order book when it's tight) and are shown for comparison only. A gap between polls and markets is a signal to look closer, not proof either side is right.
Congressional records and photos from public-domain government data; campaign finance from the FEC; bills from Congress.gov; voting-record scores from Voteview (UCLA); background from Wikipedia (CC BY-SA 4.0) and campaign sites. Positions are quoted from candidates' own sites, and every quote is checked word for word against the page. Crosstab doesn't rate candidates.
Straight news and pollster releases about tracked races, never opinion pieces. Articles about the same event are grouped; "Breaking" means three or more established outlets reported it within two hours. Headlines link to the original reporting; the line under each is Crosstab's own data.
The averages, leans, odds and ratings are calculated by the statistical model above, not by AI. Crosstab uses AI to turn those results into words: the daily studio briefing, the morning summary and social posts. The AI receives only the model's numbers, and any draft containing a number that isn't in the data is rejected and replaced.
Poll results belong to the organizations that conducted them. Poll listings are compiled with the help of RealClearPolling, Wikipedia's election pages, the VoteHub Polling API (votehub.com, CC BY 4.0) and the pollsters' own published releases. The House forecast uses The Downballot's 2024 presidential results by congressional district (the-downballot.com) and district boundaries from the U.S. Census Bureau (numbers read from a release are kept only when each one is printed next to its label in that release); Wikipedia material is used under the Creative Commons Attribution-ShareAlike 4.0 license. Betting prices: Polymarket. Everything else on Crosstab (the averages, leans, forecast, ratings and analysis) is Crosstab's own.