Ways to Predict the Outcome of 2026 Mid-Term Elections
On the 2026 mid-term elections. This post concerns various ways to predict the outcome of the election.
Featuring guest blogger, Harrison Hickman
+++++
The period between Labor Day and Election Day is filled with predictions about the election outcome. This post examines how predictions are made and what we should expect when votes are counted this year.
Changes in the balance of power in the US House of Representatives are the most common focus of macro-election predictions in off-year election years. Unlike state legislative, gubernatorial, and US Senate races, elections are held in every House district nationwide and every eligible voter in the country can participate. The results are perpetually if not correctly read as an indication of the “mood of the country.”
The “iron law” of mid-term elections is that the party of the incumbent President will lose seats. In the last 75 years, the incumbent party has lost an average of about 25 seats. Only twice has the President’s party gained seats: in 2002 in the aftermath of 9-11; and in 1998 in the aftermath of the Clinton impeachment.
Mid-term seat losses have ranged from landslide (63 seats in 2010, 54 seats in 1994, and 40 seats in 2018) to relatively trivial (5 seats in 1986, 8 seats in 1990, and 9 seats in 2024). As a result, the main question and the focus of predictions is the magnitude of losses that the President’s party will suffer.
Until recently, most predictions had one of two bases: what voters said they were going to do or predictions based on things likely to impact voter decisions whether or not voters were aware of the connection. Polls are the most obvious way to measure what voters say they intend to do. The accuracy of publicly available polling has declined as conducting polls has become more difficult and as increasingly untrained people have begun posing as “pollsters.” The result is that polls are becoming a less reliable means for predicting election outcomes, especially in volatile and low-turnout elections.
The most famous of the major alternative was a “referendum” model created in the 1970s by Yale political scientist Edward Tufte (and later updated by Michael Lewis-Beck and Charles Tien). The underlying notion was that if people are pleased with the President and the economy is doing well, the President’s party will benefit; conversely, it will suffer if the President is unpopular and the economy is not doing well. Using Presidential approval and the growth rate of personal disposable income as variables, the model has proved to be a reliable predictor of House races at least in terms of the direction of changes that occur.
Other models have been developed, and a host of analysts, commentators, thumb suckers, and pundits – in addition to random Uber drivers and distant relatives -- make predictions about what is likely to happen. But the most fundamental and most visible alternatives to polls and the referendum models are wagering and prediction markets that have arisen in recent years. These advertise that they are better predictors than polls, but much of this superior “accuracy” comes from races that do not have polls or at least polls near election day.
What should we expect this year? The one thing we know is that it would not take much to shift control of the House. Republicans currently hold a 4-seat majority.
Generic polls asking respondents which party candidate they will support in the Congressional race are showing a narrow but persistent advantage for Democrats. Based on previous poll-based predictions, this suggests a Democratic gain/Republican loss of between 13 and 26 seats. Nate Silver’s poll-based model gives Democrats an 87% probability of taking control.
Given President Trump’s low job performance ratings and tepid growth in disposable income, the “referendum” model currently predicts a Democratic gain/Republican loss of 28 seats.
The prediction markets have a similar estimate. For example, Kalshi says the probability of Democrats winning control is 85%, while Polymarket has the probability at 86%. The Polymarket seat projection is that Democrats will control the House by a margin of 232 to 203, or a 33-seat gain.
Two things could upset or at least mitigate this strong trend toward Democrats. First, Republicans have substantially more money to spend on election campaigns this year, and a recent Supreme Court decision will allow the parties to spend as much as they want in elections. Second, net changes in district lines by state legislatures will help Republicans. Larry Sabato, a political scientist and analyst based at the University of Virginia, estimates that redistricting will translate into Republicans winning 5 to 9 more House races than they would have won if the districts had not changed.
It is also important to note that losing control of the House would not mean Democratic control of Congress. Most analysts and prediction markets give Republicans a better than average chance of retaining control of the Senate even though Republicans are defending 22 of the 35 Senate seats up for grabs this year.