Our prediction model for America’s midterm elections

Also: Virginia’s redistricting may be the nail in Republicans’ coffin

Marie Segger

Data journalist and interim special-projects editor

We have launched our prediction model for America’s midterm elections. It conducts 25,001 simulations of every race, based on the latest primaries, polls, fundraising data and more to produce a statistical forecast for the House of Representatives as well as for the Senate. The recent redistricting in Virginia may be the final nail in the Republicans’ coffin: the model currently suggests that the Democrats have a 19 in 20 chance of taking the House. Republicans’ hopes rest on the Senate, but even there the race looks close: the chances are split nigh on evenly. Still, Democrats need to flip four of the 100 seats to gain control—a tall order given that all but two of their potential targets are in states that Donald Trump won by double digits in 2024. Read more about the ins-and-outs of our forecast and explore the data.

More of our data and visual reporting this week:

• Noise is often dismissed as a mere nuisance. But it can have lasting health effects. Indian cities are among the loudest across the world; noise levels in Delhi’s streets average around 75 decibels—four times the threshold recommended by the World Health Organisation. Vishnu Padmanabhan explains why India’s government is struggling to put a stop to a deafening problem.

• Mr Trump’s tariffs on China were meant to hobble its exports. But although those to America fell by more than $100bn in 2025 (a 20% decline from 2024), China’s exports to other countries increased by around $300bn. Our calculations show that Chinese companies are finding genuine new markets, not just new routes to America.

• Below Dan Rosenheck outlines why our model looks so bullish on Democratic success in America’s midterm elections in November. With more than six months left before voters cast their ballots, our claim that the Democrats are heavy favourites to flip the House might seem bold. But it is all in the data.

---

Captain Obvious: Forecasting America’s midterms

Dan Rosenheck

Data editor

Contrarian arguments are a time-tested method to attract attention. A somewhat provocative take, just plausible enough to be heard out, can often make for a successful article pitch. For the latest edition of The Economist, I took the opposite tack: I put forth what may seem like the most banal conventional wisdom possible, but emphasised that I’m really, really sure that it is right: Democrats will win the House of Representatives.

The Cook Political Report favours Democrats in 217 of the 435 seats, so barring any upsets, the party would need to win just one of the 16 “toss-up” races to secure a majority. Similarly, bettors on Kalshi, a prediction market, give the party an 86% chance of flipping the lower chamber. The Economist’s new Congressional forecast agrees—and then some. It currently puts the party’s probability of winning the House at 97.9%.

Those old enough to remember 2016—when Sam Wang, a professor at Princeton, had to eat a bug on television after a similarly confident prediction for Hillary Clinton to become America’s president went awry—may well be raising their eyebrows at such brash certainty, particularly this far out from election day. And I’m not racing to bet on the Democrats, largely because one of the model’s foundational assumptions will probably not hold: its simulations are based on current district boundaries, which are likely to change as a result of new Republican gerrymanders. (In fact, Florida’s legislature is set to discuss new maps today.) But if the current maps are used, a Democratic House in 2027 is as close to a sure thing as modern American elections can offer.

The biggest reason for our model’s confidence that this will be the Democrats’ midterm is that, well, it’s a midterm. The president’s party almost invariably loses popularity during the first two years after taking office, and “generic-ballot” polls—which ask respondents which party they plan to support for Congress—suggest that this historical pattern persists. Among decided voters, Democrats lead by around 53% to 47%.

Although there is plenty of time for the polls to shift, they haven’t tended to move all that much in modern midterms. Starting in 1994, the election that ended Democrats’ 40-year-long hold on the House, the president’s party’s two-party vote share on the generic ballot has changed by an average of just 1.7 percentage points during the final 190 days of midterm campaigns. And that movement has generally been against the party in power: the average shift has been 1.3 points towards the “out-party”. By contrast, in presidential-election years, the “in-party” has gained 0.6 points on average. Our model accounts for this pattern, projecting that the current polling average will inch towards the long-run midterm average by election day.

---

The president’s party tends to fall in polling averages during the campaign. Our model incorporates this pattern into its forecast

---

Even once the campaign reaches its end, the generic ballot can misfire: in 2020 it overstated Democrats’ two-party share of the national House vote by three percentage points. To reduce this risk our model incorporates other sources of information, such as the results of special elections for vacant legislative seats and the president’s approval rating. Looking at all midterms since 1942, our model’s estimates with 190 days to go have missed the mark by an average of 1.4 percentage points of vote share. That is far worse than the 0.9 points it achieves on election day, but less than half of the Democrats’ current lead.

---

Even with 190 days to go, our model can predict the popular vote in midterm elections with reasonable accuracy

---

Winning the popular vote does not guarantee a majority of seats. But because the current House-district boundaries are almost perfectly fair nationwide (though highly unfair within gerrymandered states), whichever party wins more votes is also favoured to win more seats. And in the past 40 years, the only time the president’s party won the popular vote in a midterm was 2002, in the wake of the September 11th attacks. Given Donald Trump’s abysmal approval rating, he is highly unlikely to become the second. How unlikely? Well, the chances are around 2%.

---

Virginia’s redistricting may be the nail in Republicans’ coffin

On current maps, our model sees Democrats as all but certain to flip the House

The sun rises in the east; the other queue always moves faster; and the president’s party loses the House of Representatives in the midterms. Starting with the Democrats’ blue wave in George W. Bush’s second term, this pattern has held firm every four years. Even Republicans’ weak performance in 2022, derided as a “red ripple”, ended Joe Biden’s trifecta.