Every cycle the polls differ to some extent from the election results. It's inevitable when pollsters can only give estimates about who will show up to vote, some people don't make their decision until they're in the voting booth, and bombshells can happen late in the race.

Data from the past four decades shows that the polls do not always bias one party over the other and that past results cannot predict how the polls will perform next time. The 2022 midterm election polls, for example, were among the most accurate in years.

To assess the accuracy of presidential polls, the charts in this article show averages that combine many polls into one estimate for each election. Between 1988 and 2020, the final national polling average was off by an average of 2.3 percentage points.

Some years have been better than others: In 2008, the national polling average missed Barack Obama's bottom line by an average of less than a percentage point; In 1996, she overestimated support for Bill Clinton by almost four points.

State-level polls haven't fared quite as well. Since 2000, polls in nearby states have been off by an average of 3.1 points. In 2016 and 2020, almost all state polling averages underestimated support for Mr. Trump, sometimes significantly.

In the last two presidential elections – two very close races that have increased the importance of the Electoral College – voting errors in the states became even greater. But the results of the polls in these two years were not entirely unusual. A look at state polling errors since 2000 shows that polls in previous elections performed about as well as polls today.

State polls have missed in both directions over the years. But if pollsters underestimated Trump in the last two elections, are they doomed to do so again this year? Should you, as some poll watchers claim, mentally add a boost to Mr. Trump to all the poll numbers you see?

Pollsters are convinced that they have largely identified the causes of the 2016 election failure. A lack of consideration of voters' educational levels was a key reason, according to a report by a professional public pollsters organization. Particularly in surveys at the state level this year, respondents with a college degree were overrepresented and respondents without a college degree were undercounted.

In previous elections this was less of a problem because voting decisions did not differ so much based on educational aspects. But in 2016 and beyond, voters without college degrees have largely supported Republicans, particularly Mr. Trump.

By 2020, almost all pollsters had started taking education into account. But polls still underestimated Mr. Trump. This time the cause of the error was less clear. One theory, laid out in a 2020 poll analysis report, is that Trump supporters were less likely to participate in polls. As a result, “those who respond to the poll, even when accounting for non-college white men, are more Democratic than those who don't,” said Chris Jackson, who leads U.S. public opinion research for Ipsos. Others have posited that Biden voters were more likely to stay home during the pandemic, giving them more time and opportunity to participate in polls.

Another challenge this year: record participation. The Pew Research Center estimates that about a quarter of 2020 voters had not voted in 2016. And polls had shown that any new voters in 2020 would be overwhelmingly Biden supporters. In fact, according to the Pew study, they were split between Mr. Biden and Mr. Trump.

So what about this year? The polls in seven swing states are extremely close – in most of these states, Trump and Kamala Harris are essentially tied. While the polls in these states underestimated support for Mr. Trump in the last two cycles, they have historically had a mixed track record, with failures on both the left and the right, and some years have been better than others.