Polls have been predicting a Democratic surge, but last-minute deciders and sampling issues mean there's no guarantee. By Anthony Zurcher The Democratic Party has high expectations as it heads into these pivotal US midterm elections.

A historically unpopular Republican president, an American public unhappy with the rising costs of food and bills, and months of favourable polls point to a good night for them on Tuesday.

Democratic candidates lead in well over the 218 seats needed to flip control of the lower chamber of Congress, the House of Representatives, the polls suggest. And they are tied or narrowly ahead in the key Senate races.

A recent New York Times poll had the Senate race in South Carolina, a state Donald Trump won by 18 percentage points in 2024, as a tie. Recent CNN surveys put the Democrats ahead in Texas, Kansas and Maine. But are the polls accurate?

Recent US political history is littered with examples where public opinion polls, even ones just days before an election, were off the mark. Trump outperformed predictions on his way to White House victories in 2016 and 2024. His narrow defeat in 2020 came after most surveys indicated a more comfortable win for Democrat Joe Biden.

Earlier this year, polls suggested Democrat Abdul El-Sayed was poised for a double-digit win in Michigan's Democratic Senate primary. Instead it was 1%. In Wisconsin's Democratic primary for governor, the candidate that had a polling lead that approached 20 points lost by a fraction of a percent.

According to Clifford Young, president of Ipsos Public Affairs, which conducts political and market research surveys, polling has become difficult in recent years because it's harder to reach Americans as they abandon their land lines and screen their calls. And many of those who pollsters do contact have become less willing to participate. "Conservatives, especially those on the far right, are much more anti-establishment by nature," Young says. "And they don't participate in anything. They hate the system. That's the challenge."

If Republican voters are being undercounted in a systematic way, and polling companies don't take that into consideration when constructing their voting models, it could skew their data towards the Democrats. That's one of the reasons why it has become trickier for pollsters to predict what the electorate will look like.

GUIDE: What are the US midterms? IN DEPTH: The real reasons there are so few young US politicians EXPLORE: Where the US midterms will be won and lost VERIFY: How the US economy is really doing under Trump One explanation for why Trump overperformed in previous presidential elections is that his supporters – many of whom were not regular voters – flocked to the polls in higher-than-expected numbers.

That unpredictability is one of the main contributing factors to those notable recent polling misses, even if the margins in most reputable polls were only off by a few percentage points.

In a close race, those few points can make the difference between victory and defeat for a candidate. And this year, many of the key Senate races are within those narrow margins.

In recent years, midterm general election polls have proven to be more accurate, in part because Trump - and his unique ability to draw disaffected Americans to the polls - was not on the ballot.

In 2022, they suggested Democrats would defy conventional expectations and avoid an electoral wipe-out despite Biden's low approval numbers.

While the party lost control of the House of Representatives, the margins were slim and Democrats maintained control of the US Senate.

While this year's primary results did provide bigger surprises, Young says that nominating contests have historically been volatile, with voters more likely to make up their minds at the last minute. It shouldn't be seen as an indication that the current crop of general election polls are off the mark. It's common for the party of the president to have a bad night in the midterms.

One of the most notable examples in recent years came in 2010 when Republicans flipped more than 60 House seats and made gains in the Senate.

It came off the back of an economy struggling to recover from the financial crash and conservative fears over President Barack Obama's healthcare reforms. A defeat had been predicted by the polls but not quite on that scale.

The roles are reversed this time, with the Democrats in opposition and the economy on the minds of voters. But there are a number of reasons why Democrats may not have as good a night as history might suggest or as some might expect, given Trump's low approval rating.

Firstly, much of the territory in play in this election Trump conquered with ease in the presidential election just two years ago. It's like a home game for Republicans.

Then there have been changes to House election boundaries over the years that have meant fewer seats switch between parties.

Thirdly, there are fewer swing voters due to the deeper polarisation of the country. Some experts describe the electorate as calcified.

There's also the cash advantage Republicans have enjoyed which has given them hundreds of millions of dollars to spend on attack ads in the final weeks.

Polling models try to capture shifting fundamentals. And a well-designed poll is built to anticipate issues that may make it less than reliable.

Individual polls do have a built-in margin of error – a statistical level of uncertainty based on the sampling size compared to the population as a whole. Taken as an aggregate, polling averages and not individual surveys can provide a more accurate prediction.

At the moment, even the kind of polling lean we saw in recent presidential elections would suggest Democrats are likely to have a good night on 3 November. And polling inaccuracies can go both ways. In last year's state and special elections, Democrats outperformed most surveys as it was their voters who turned out in unexpectedly larger numbers.

The sight of Trump holding rallies in the final days of the campaign in states he won comfortably in 2024 will have lifted Democrats' hopes that Republican internal polling has conservative candidates in trouble. There's always the chance that voters change their minds right up until election day, however.

Sometimes there are late shifts that can alter the dynamics of a race. Sometimes undecided voters, which still account for a key segment of the electorate in many of these races, all break the same way.

In polling, there will always be some level of uncertainty – and managing that is the art that goes along with the science.

"I think we're doing a better job," says Young. "But part of the problem we always have is not just with the pollsters. Part of it is that we're dealing with humans. Humans are not easy." Follow the twists and turns of Trump's second term with North America correspondent Anthony Zurcher's US Politics Unspun newsletter. Readers in the UK can sign up here. Those outside the UK can sign up here.