Opinion polls put the Democratic Party on course for a major victory in the 2026 US midterm elections, less than a month before polling day, with its candidates ahead even in states that strongly backed President Donald Trump in the 2024 election.
Democrats forecast a large House majority
Current projections suggest Democrats could secure their largest House majority in 18 years, but the polling record in previous election cycles calls for caution before deciding the outcome prematurely.
An analysis by writer Mark Novikov warns against placing too much confidence in these figures, noting that people working in the polling industry have repeatedly believed they had addressed shortcomings in their models, only for subsequent elections to reveal that errors persisted or that new problems had emerged in measuring voter sentiment.
Polling errors after the 2016 election
After the 2016 election, pollsters concluded that they had not given enough weight in their samples to voters without university degrees, after educational attainment became an influential factor in political polarisation. Polling firms introduced adjustments to address the problem, helping produce relatively accurate results in the 2018 midterm elections.
But the problem returned in the 2020 presidential election, when polls again underestimated support for Trump, particularly in their state-level estimates. In Wisconsin, some polls projected that Joe Biden would win by as much as 10 points, while his actual victory margin was less than one point.
Following those results, pollsters concluded that Trump supporters were less willing than others to respond to polling firms. This explanation may help clarify why midterm election polls, in which Trump is not a presidential candidate, have generally been more accurate than polls for the presidential elections in which he ran.
New models for measuring support for Trump
To address the problem, some firms adopted a method known as “vote recall”, which involves asking respondents which candidate they voted for in the previous election and then adjusting the sample composition to reflect the share who voted for Trump. The method produced successful results in 2022, but it did not prevent polls from underestimating Trump's support by about 3% in the 2024 election.
These difficulties prompted polling firms to develop more complex models, including analysis of electoral neighbourhoods and voters' differing characteristics. The New York Times and Siena College also developed new statistical tools to adjust samples based on a range of indicators, rather than relying on a single variable.
The impact of turnout models on the electoral margin
The problem is not limited to identifying which candidate a voter will choose; it also involves estimating who will actually turn out to vote. The impact of turnout models can be seen in a YouGov poll, in which Democrats led by six points among registered voters before the margin rose to 16 points when those classified by the poll as likely voters were counted.
The nature of the 2026 election may lend some credibility to current projections, as it is a midterm election in which Trump's name will not appear on the ballot. This could reduce turnout among voters who are less interested in politics, a pattern that midterm polls have previously handled with greater accuracy. At the same time, the sample of elections held during the Trump era remains limited.
The president has also sought to keep the race centred on him by selecting candidates, promoting the Republican Party and turning party conventions into events focused on him and his political presence. Novikov warns that a repeat of an error on the scale of those seen in 2016 or 2024 could deprive Democrats of control of the Senate.
If the error reaches the level seen in the 2020 election, the party may make no net gains in the chamber, while regaining the House by a narrower-than-expected margin. These scenarios remain hypothetical, as the size of polling errors can vary from state to state.
