The stakes are considerable. The elections on 3 November 2026 will contest all 435 seats in the House of Representatives and 35 Senate seats, and control of Congress could hinge on a handful of highly competitive races. Dr John Kenny, co-Guest Editor of the special issue from the University of Mannheim and University of Southampton, noted that midterm elections have historically seen the president’s party lose support, driven by economic concerns and falling approval ratings. But the political landscape of 2026, he argued, is atypical. Increased polarisation and widespread redistricting have introduced new uncertainties, complicating the question of whether the election will follow established patterns or mark a significant shift in American politics.

Dr Debra Leiter, co-Guest Editor from the University of Missouri-Kansas City, framed the asymmetry between the two chambers in structural terms. With such a narrow Republican majority, Democrats require only modest gains to retake the House, but the Senate is harder to flip because of the number of Republican-held seats considered secure. She pointed to record-breaking spending, open-seat contests created by a wave of retirements, and unexpected candidate choices as forces expected to increase competitiveness across key battleground states. Coupled with the unique circumstances of the 2026 election, she concluded, it is a tough one to call, raising the question of whether the fundamentals of American election forecasting will reassert themselves or whether the cycle will challenge widespread beliefs about American electoral behaviour.

Among the most technologically ambitious contributions is Roberto Cerina’s exploration of whether AI-generated synthetic personas can replicate human respondents in pre-election polling. Working at the University of Amsterdam, Cerina constructed detailed large language model personas based on factors such as age, gender, education and political identity. The model’s key innovation is its capacity to simulate media consumption: using up-to-date news feeds, the personas were exposed to current events and asked how those developments influenced their political preferences. This design helps overcome a well-known limitation of conventional AI models, whose knowledge is frozen at training cut-off dates. The silicon sample forecasts a Democratic victory in the House, with 58.4 per cent of the national vote and around 292 seats. Cerina acknowledges the forecast may overestimate Democratic support, but argues the approach offers a transparent way to understand why voters change their preferences, positioning AI polling as a complement to traditional surveys that could deliver more frequent and granular insight into public opinion.

A very different tradition of forecasting comes from Thomas Gruca and Thomas Rietz, who compared the Iowa Electronic Market, a classic futures election market, with Kalshi, one of the fastest-growing online betting platforms. Their analysis demonstrates the stability of IEM forecasts in contrast to the greater instability of the Kalshi markets. In separate markets, the IEM predicts that the most likely outcomes are a Democratic House victory and a Republican Senate majority, while Kalshi markets predict a Democratic House and Senate majority, though less consistently. In combined control markets, both indicate that the most likely joint outcome is a Democratic sweep. The most recent IEM prediction gave Democrats a 60 per cent chance of taking the House, whereas Kalshi put that probability significantly higher, at close to 90 per cent. Both markets ultimately predict a Democratic Senate. The authors note that the IEM has an established history of effective and reliable forecasts that may be less prone to certain kinds of market manipulation.

District-level granularity anchors the contribution from Peter Enns, Leigh Farah, Claudia Miner and Thomas Gareau-Paquette of Cornell University. They observe that while observers focus on who wins the majority, 435 separate district contests will determine political authority in the House. Their model incorporates key election factors in each district, including this year’s unique mid-decade redistricting, candidate success measures such as fundraising, and both national and state-level vote intention, blending national and localised information in a new way. Based on 10,000 simulations using data from 100 days before the election, the forecast gives Democrats a 90 per cent chance of taking the House, with an expected total of 235 seats.

The long-running Political Economy Model, unveiled in updated form by Charles Tien of Hunter College, CUNY, and Michael Lewis-Beck of the University of Iowa, rests on three key predictors: economic performance, presidential approval ratings and the historical impact of midterm elections. The model’s logic is that voters reward governing parties when conditions are favourable and punish them when they are not. With President Donald Trump’s approval rating at around 35 per cent, the model predicts Republicans will lose 33 House seats and their narrow majority, and that Democrats will pick up enough seats to retake the Senate. The forecast reflects the powerful historical tendency for presidents’ parties to suffer losses at midterms.

History on a longer timescale informs the model developed by Stephen Quinlan and Michael Lewis-Beck, which examines the structural and historical conditions that have consistently influenced election outcomes over decades. When one party controls the presidency, Senate and House simultaneously, voters have tended to impose an electoral penalty on the governing party. With Republicans currently controlling all three branches, conditions appear favourable for Democratic gains, although increasing political polarisation and a declining number of competitive House districts may limit them. The central forecast projects Democrats winning 231 House seats and 50 Senate seats, translating into a 78 per cent chance of recapturing the House and a 66 per cent chance of winning control of the Senate.

Uncertainty itself becomes the subject in the study by Jack Rametta and Christopher Hare of UC Davis, who invoke the Rashomon effect to show how model choice can obscure forecast uncertainty in the 2026 Senate picture. Their model reflects the growing nationalisation of congressional elections, in which local candidate appeal and the incumbency advantage are increasingly outweighed by broader forces such as presidential approval, partisanship and economic conditions. Elections once shaped more heavily by state-level dynamics are now closely tied to voters’ perceptions of national political performance, meaning even well-regarded incumbents can struggle against unfavourable national trends. As of late September, their forecast predicts Democrats will take 50 Senate seats, but with high uncertainty the range spans 45 to 56 seats, leaving Democrats slightly more likely than not to secure a majority in what remains a highly competitive race.

A practical problem confronts many forecasting models this cycle: the discontinuation of Gallup’s Presidential Approval polls, an 88-year tradition. Olivia Quinn, Amanda Brush and Eric R. A. N. Smith of UC Santa Barbara show that models can continue without it by developing a proxy measure from RealClearPolitics’ poll average. As of the most recent quarter, their Congressional seat change forecast predicts sizable Democratic gains in both chambers, with Republicans losing nine seats in the Senate and 42 in the House. Taken together, the seventeen studies sketch an election in which the House leans decisively toward Democrats across nearly every methodological lens, while the Senate remains genuinely contested, a split verdict that may itself be the most honest summary of a polarised and rapidly redistricted American political landscape.

Subject of Research: Forecasting models for the 2026 US midterm elections

Article Title: Forecasters predict Democrats to take House but race for Senate hard to call

Article References: Forecasters predict Democrats to take House but race for Senate hard to call. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: 2026 midterms, election forecasting, House of Representatives, Senate, political economy model, prediction markets, synthetic personas, polling, redistricting, presidential approval, polarisation, Cambridge University Press