US presidential election 2020: Are the polls tightening?

Outlandish predictions always attract attention online, but changes over recent days appear to be mostly noise.

Sign Up

Get the New Statesman's Morning Call email.

Have you got it? No, not Covid; pre-election anxiety. Many on the Democratic side – whether candidates, canvassers or supporters around the world – are wondering whether their confidence about winning the upcoming US presidential election might be misplaced. Nerves, made worse after the poll miss four years ago, are reaching a peak as election day looms. Be it wide margins for Donald Trump in Iowa, or a tightening race in Pennsylvania, it seems there has never been so much attention paid to every last detail of the final polls.

And yet Joe Biden has a lead almost unheard of in recent times. No candidate over the course of the last five election cycles has led an opponent by as much as Biden, nor for as long. 

In the states where it matters, such as Pennsylvania, Arizona and Michigan, Biden has a three- to seven-point lead. Even if we apply the poll error of 2016 to the numbers we’re seeing today, Biden would still become the 46th president of the United States. 

For Trump to win the electoral college on current numbers, pollsters would have to understate his support by more than they did in 2016. That is unlikely: since that miss, pollsters have made notable methodology changes, with more weight put on factors associated with support for the Republican candidate, such as education (those without a college degree) and ethnicity (white).

What would happen if the polls were as wrong as in 2016?
Joe Biden's poll lead readjusted if pollsters were to understate Donald Trump's vote share by the same margin as in 2016

In testing those changes, most pollsters predicted the vote shares for the 2018 midterms with reasonable accuracy (the Democrats gained the house with 53 per cent of the vote). The notable exception was the pro-Republican pollster Rasmussen which, in its final poll, understated the Democratic share by nearly ten points. Since 2018, I am not aware that they have sought to improve their methodology; their latest survey gives Joe Biden a lead of just three points.

That said, Trump is an enigma. How much of his vote is his vote? Could some of that vote be less enthused to vote in midterm elections when Trump’s name, despite efforts by opposing campaigns to make it so, is not on the ballot? Perhaps, but there are still the presidential polls, and in those Trump is trailing Biden by notable numbers.

Why, then, is there so much anxiety? Regrettably, most voters remember failure more than they do success. It is in part why the implied probabilities of betting odds have always presented this race as tighter than forecast models indicate. 

[See also: Why I’m starting to doubt my long-held belief that Trump will win the US election]

But we're not here to talk about betting odds. In the last few days we have seen pollsters publish their “final polls”. Some have offered margins that are much changed from a few days prior, and others have detected next to no shift in voting intention. For the most part, we've seen no notable difference between now and last week, and what changes exist appear to be noise. The latest margins from the New Statesman’s US election model give Biden comfortable leads in all of the states he needs to win the White House.

Have the polls changed?
Joe Biden's poll lead by state, taken every Monday since 28 Sep

A couple of market researchers have sought to introduce 11th-hour methodology changes. Some have chosen to up their sample size to minimise error. Others have introduced new ways to forecast turnout and separate voters who have already cast their ballots – now a considerable figure. As of 28 October, 54 per cent of the total number of Americans who voted in 2016 had already cast their votes early. This has only surged since.

There is a school of thought, be it from Nate Silver or other analysts in the data world, which considers that pollsters aren’t doing enough, or are doing too much, to exclude those who have already voted from their polls, and in doing so risk producing phantom shifts in voting intention. Republicans will likely lead on votes cast on the day, whereas mail-in ballots will favour Democrats. A potential tightening in the polls may be a consequence of this methodology change, and also just that: a phantom shift.

Pollsters are, like any other business, conscious that they need to win clients and conscious, too, that they need to appear “right”. When their final figures in a high-stakes election are often the figures analysed and considered by potential clients, it’s understandable that methodologies may be altered or samples upped.

But trying to be the one “right” source is risky. For some researchers, it’s better to be among the pack of pollsters, closer to the average rather than the extreme. That conservatism leads to herding –pollsters pushing their final figures towards an average that, if proven wrong, doesn’t isolate just that single company as being in error, but the industry as a whole. 

With that in mind, my advice is this: don’t pay as much attention to the final polls as social media wants you to. The more outlandish or tighter the poll, the more likes and retweets it gets on social media. Instead, follow the poll trackers and models that weight their numbers with fluctuations in mind. It is in part because of potential error and the risks of herding that, unlike others, the New Statesman’s tracker still gives Donald Trump a 10 per cent chance of winning re-election.

Do not overstate or understate that probability: a one in ten chance is unlikely – very unlikely – but still well within the realm of possibility.

[See also: The New Statesman's presidential election poll tracker]

 Ben Walker is a data journalist at the New Statesman

Free trial CSS