The national environment · Updated
Before forecasting any single race, the model estimates the national mood: how the House popular vote will split nationwide. Our estimate is
Three readings, one estimate
Each reading is corrected for how it has missed in past elections, then weighted by how accurate it has been. The shaded band is where the result has landed 80% of the time given that reading: the wider and fainter the band, the less reliable the reading, and the less it counts.
The generic ballot
Each dot is a poll, shown as the two-party margin (undecided voters set aside). The line is the average the model uses, with each poll's weight halving every two weeks. The model then subtracts about three points, the amount September generic-ballot polls have overstated Democrats in 14 of the last 15 elections.
Special elections: the turnout signal
Every special election contested by both parties since 2017. Each dot shows how much better (up) or worse (down) the Democrat did than the district's last presidential result. These low-turnout races reveal which party's voters are more motivated.
Presidential approval
President Trump's job approval. Each faded dot is a poll; the lines are the average, with each poll's weight halving every two weeks. Polls from Decision Desk HQ and VoteHub. Approval feeds the fundamentals reading, which gets the least weight because it has been the least accurate predictor on its own.
Every election since 1978
Polling lines start in 1996. In 14 of the 15 elections since, the late-September polls were more Democratic than the final result (the black line sits above the bar's end), by 3.4 points on average. That's why the model corrects for it.