Last week’s polls got the US elections wrong – but not THAT wrong

How bad were the American polls last week? Until the final figures come in, any verdict must be provisional. However, at the time of writing, this seems to be the picture.

  • The polls systematically understated Donald Trump’s support.
  • This meant that the predicted landslide (in electoral college terms) for Biden didn’t happen – and nor did Democrat victory in the Senate or the expected gains in the House of Representatives.
  • The polls did worse than in 2016, where the national polling averages came close to the overall result; the main errors four years ago were limited to just a few states – though these happened to be the ones that delivered victory to Trump. This time, Trump beat the polling predictions both nationally and in most battleground states (but not all: they came close in Georgia and Minnesota).


  • The polls’ errors, though politically serious, were no further adrift than in many past elections. Taking the two-party share of the vote, the polls on average predicted a national victory for Biden by 54-46%. When the final figures come in (there are millions of votes still to be counted in strongly Democratic California and New York), Biden’s victory in the popular vote is likely to be 52-48%. Nationally, the polls were systematically wrong but not hugely wrong.
  • Assuming that Biden stays ahead in Arizona and Georgia, the polls predicted the right winner in 48 of the 50 states. To be sure, a number of them greatly overstated Biden’s lead in the battleground states; but the only two states they called wrong were Florida and North Carolina; but even there the final average polling leads for Trump (2.5% and 1.8% respectively) made both states too close to call. Predictions of a Biden landslide always rested on shaky foundations.
  • Much has been made of the Democrats falling short in both the House and the Senate. Did millions of Americans vote for Republican candidates in these races while rejecting Trump? Maybe in some cases, but not generally. In 2018 the Democrats enjoyed a national lead of 8% in the popular vote in elections to the House. It will be around half that this time. In short, there has been a nationwide pro-Democrat swing since 2016 in the election for president – but a pro-Republican swing since 2018 in elections for the House. That explains the apparent contradiction between the Democrats winning the White House but losing ground in the House of Representatives.
  • One curiosity: in the key states of Michigan, Wisconsin, Pennsylvania, Florida and Texas, the final polls from 2016 proved better predictors of last week’s results than last week’s polls. Likewise the national polls: Biden’s four-point lead in the popular vote was exactly the final polling average four years ago. Make of that what you will.

We shall have to wait for a full post mortem to find out why the polls were adrift this time. AAPOR, the American Association for Public Opinion Research, did a thorough job last time. It’s likely to hold another inquiry.

Meanwhile, my guess – and it’s no more than a guess – is that some Trump supporters were so alienated from mainstream institutions, ranging from TV networks to pollsters, that they declined to join online polling panels or respond to telephone calls soliciting their views. But we’ll see.

There is, though, a wider issue. With polls – and they are not alone in the realms of political discussion – public discourse seems to acknowledge only two verdicts: right or wrong (or, for Trump, fake). Having been a pollster, I know only too well that the truth is usually different. Polls are seldom precisely right, and seldom totally wrong. They generally come close to the truth, but not always close enough.

We should, then, reject crude, binary judgements. Studied sensibly, polls tell us much, but not everything, about the way citizens view their own lives and the actions of those with power and money. Without polling data, our understanding of our democracy would be much weaker. For all their faults, polls remain our least bad way of understanding public opinion.

This blog was first published by The Article