America’s polls appear to paint a picture of doom for Trump. The Democrats are on course for a clear victory in the House of Representatives, but Republicans have an outside chance of retaining the Senate. That would be a significant consolation for Trump, who would be better able to divide and rule for the rest of his presidency.

Two sets of polls published on Sunday illustrate the point. YouGov has the Democrats ahead in seven key states, three of them very narrowly: by just one point in Kansas and Iowa, and two in Ohio. In polling terms, tiny Republican majorities could reflect sampling bad luck. YouGov could not necessarily be blamed if the Democrats and Republicans both end up with 50 seats, and Vice President JD Vance’s casting vote kept the Republicans in control. 

The New York Times/Siena polls also show the Democrats ahead in the battleground states – but with one exception by bigger margins. They have the Democrats leading by 5 to 11 points in Kansas, Maine, Michigan, Ohio and Texas. Assuming that the Republicans lose North Carolina (which all pollsters agree is a certainty), the Democrats should end up with at least 52 Senators, and a clear majority. 

On the basis of the final polls, the likeliest outcome is a 52-48 majority for the Democrats in the new Senate. But we are dealing with probabilities, not certainties. It is certainly possible that the final outcome could vary by two seats in either direction – either a 50-50 split or a Democratic sweep of the battleground states, giving them a 54-46 majority.

Such modest differences, both within range of at least some of the recent polls, mark the outer boundaries of battle for the Senate, which range from inflicting disaster on Trump to giving him bragging rights for beating his critics’ expectations.

For the rest of us, the proliferation of US polls in the mid-term elections is apt to cause confusion. Is the Democrats’ lead a massive 14 points, as Atlas Intel report, or just four points, the judgement of the latest Harvard-Harris poll? 

One of the common mistakes that commentators make is to discuss the margin of error of a collection of polls as if they were a single poll. According to RealClearPolling, the polling average going into election day is that the Democrats are leading the Republicans by nine points – 52-43 per cent – in the nationwide race for Congress.  

If these were the figures from a single poll, the correct interpretation is that the Democrats are probably well ahead but that we could not be certain. Random sampling fluctuations could mean that the poll overstated the Democrats and understated the Republicans, both by three points. If so, the true figures would be 49-46 per cent. Given the advantages gained by the Republicans in altering the boundaries of Congressional districts, the outcome would be too close to call. 

But that calculation does not apply to the average of all recent polls. If lots of polls err in the same direction, that’s not bad luck but flawed methodology. As statisticians would expect, their individual figures for the Democrats vary, with most in the range of 7-11 per cent, and a few outside that range. They should be on course for a comfortable majority in the House of Representative. 

What if that is not what happens? What if Donald Trump celebrates a Republican victory on Wednesday morning? Then the polls will be in trouble. Their error would be systematic, not random. 

It has happened before. They understated Trump’s support in 2024, and failed to detect his strength in the rust belt states that gave him victory in 2016. After Trump’s first victory, America’s pollsters got together to analyse what had gone wrong. One error they admitted was their failure to take account of voters’ education. The outcome: they had polled too many Clinton-supporting graduates and too few Trump-voting non-graduates. 

They corrected for this – but still understated Trump’s support in 2020 and 2024. One possibility is that Trump continued to attract Americans who had not normally voted at all in the past, some of whom distrusted establishment institutions – such as pollsters – and kept out of their way. Back in the 1990s, British polls were caught out by “shy Tories” and underestimated the Conservative vote share. Their more recent US equivalent could be Trump refuseniks. 

This theory has the advantage of explaining why the pollsters did so well in the mid-term elections in 2018 and 2022. Trump was not on the ballot paper, turnout was (as usual) lower than in presidential elections, as those pesky Trump refuseniks stayed at home. 

Overall, the Democrats can be confident that they will gain control of the House of Representatives. If they don’t, it means that more than 20 different polling companies, each separately doing their best to get the result right, have done something badly wrong. I remember the pressure to get British elections right when I headed YouGov’s political team. Of course pride came into it; but I was keen to protect YouGov’s reputation for accuracy. American pollsters will be feeling the same pressure this week. 

I would be surprised – no, astonished – if the Republicans retain the House. The sheer volume of national polling data showing big Democratic leads makes that improbable. But if it does happen, the crisis in polling would be deeper than in any previous election in the Trump era.

There is more room for uncertainty in polls of Senate races. Samples tend to be smaller, which means random errors can be larger. And while each of the main Senate races has been polled extensively, we do not have as much data as for Congress. Hence the possibility of a 50-50 division.

However, NYT/Siena pollsters will not be able to blame sampling bad luck for a failure by the Democrats to capture the Senate. Precisely because they polled seven different states simultaneously, to err so much in the same direction in all or almost all of them would mean they have got something wrong.  

Like quite a few pollsters, they have a strong professional interest in being proved right by big Democratic victories.