Trump and Harris are both within a normal polling error of a bust

In the final days before the 2020 presidential election, polls generally pointed to a clear victory for current President Joe Biden. But when the votes were counted, it turned out that the polls had overestimated him – Biden won, but narrowly. This was, of course, similar to 2016, when former President Donald Trump significantly outperformed the polls in Michigan, Pennsylvania and Wisconsin and won a surprise victory in the Electoral College despite relatively accurate national polls.

This raises two questions for 2024: First, what would happen if the polls were closed again? And secondly, how likely is it that the polls will be as wrong as they were in 2016 or 2020?

Either Trump or Harris could win comfortably

In 2020, polls overestimated Biden's lead over Trump in competitive states by about 4 percentage points. As of Oct. 30 at 11:30 a.m. Eastern Time, the polling average gap of 538 between Vice President Kamala Harris and Trump is less than 4 points in seven states: the familiar septet of Arizona, Georgia, Michigan, Nevada, North Carolina and Pennsylvania and Wisconsin. That means if the 2020 election error repeats itself, Trump would win all seven swing states and 312 Electoral College votes.

Of course, it won't necessarily help Trump if the polls are bad. It is impossible to predict the direction of poll error in advance, and polls have often overestimated Republicans in the past. In a scenario in which the polls overestimate Trump's lead by four points in every state, Harris would win all seven swing states and 319 electoral votes.

Note that none of these results are particularly close, at least not in the Electoral College.

Some polling error is normal

Both results – and everything in between – are on the table next week. But are these scenarios actually likely or are they more external possibilities? Well, this is where the work we do for our election prediction model can be helpful. As of October 30 at 11:30 a.m. Eastern time, our forecast gives Trump a 52 in 100 chance of winning the White House and Harris a nearly identical 48 in 100 chance. The model arrives at this probability by calculates how many Electoral College votes each candidate would win if certain polling errors were in their favor, and then counts how often each candidate wins in these simulations. (More on this in our methodology.)

Based on how many polls have been wrong in the past, our election model estimates that the average poll error in competitive states this year will be 3.8 percentage points.* This error is not uniform across all states – for example, states with different demographics poll error tend to have different dimensions. However, when polls overestimate a candidate, they tend to overestimate them overall. In other words, the model expects a polling error on the order of about 2020 – although not necessarily in the same direction as 2020. (In 50 percent of the model's simulations, Trump outperforms his poll numbers, and in 50 percent of the cases even Harris outperforms.)

This point is worth thinking about. Since our average expectation is that there will be a fairly large polling error at least half of the time, the likelihood that the polls will be perfect and the election will go exactly as the polls suggest is actually very small. Let's look at this through the largest lead any candidate has in the seven swing states: Trump's current 2-point lead in Arizona. Nationally, our model assumes polling error is more than 2 points in either direction 62 percent of the time. In other words, there is only about a 1 in 3 chance that polls will be off by less than 2 points (which in the past we would consider a small polling error).

Because all seven key swing states are so close together, even small voting errors in the same direction can have a big impact on who wins the election. According to our model's simulations, there is a 60 in 100 chance that either candidate will win more than 300 votes in the Electoral College – which Harris could achieve by winning five of the seven swing states and Trump winning six of the seven. By modern standards, I think it's fair to consider this a landslide victory – considering how divided the country is, it's relatively unlikely that either candidate will win much more (even to get to 320 electoral votes, Trump would have to win a state like Minnesota and Harris would have to win a state like Florida).

Of course, the likelihood of failure in both cases depends heavily on the outcome of the referendum. If Harris wins the national popular vote by 3 points, she is much more likely to win the states that decide the Electoral College than if she loses the popular vote by 3 points. This is clearly shown in the table below, which contains all the simulations in our model and puts them in the list according to the result of the popular vote.

As you can see, Trump is likely to win the election even if he loses the popular vote by 1-2 points, which is what our current national polling average suggests. And if he turns out to be underestimated in the national polls and Trump wins the popular vote by 1 to 2 points, he would be the favorite to win by a clear margin.

Meanwhile, our model assumes that Harris needs to win the popular vote by 2.1 points to be considered the favorite to win the election, as swing states are more Republican-leaning than the nation as a whole. And if she wins the popular vote by 4.5 points (Biden's margin in the popular vote in 2020), she is expected to clinch a victory herself.

Polls are inherently uncertain. That's why we model.

So far we have said little about the actual polls themselves. And there is actually reason to believe that polls this year could be more accurate than in the past. While the share of polls conducted or sponsored by Republican-allied organizations has increased – which we've written about – the overall share of partisan polls is lower than in previous years, and the average pollster rating of polls in 2024, at 538, is higher than previous ones years. . All else being equal, this should result in better polls than 2016 and 2020. We've also seen fewer polls in these years from the companies that have overestimated Democrats the most.

However, the news is not all good. In particular, pollsters still report difficulty reaching voters in the first place, and Trump supporters may still be less likely to respond to polls—even high-quality polls. This means that pollsters still (or perhaps even more!) rely on weighting and modeling to get good estimates of public opinion. But the choices they make matter a lot, and in particular there appear to be big differences between polls that try to use these techniques to balance their samples by party or past vote and those that don't.

And that's the big, fundamental problem with pre-election polls: We don't know what the demographic and political makeup of the actual electorate will be, so pollsters only make the best estimates possible. These assumptions have always been and will always be flawed.

And this is where elective models like the 538 become really helpful. Building election prediction models is not about providing a highly accurate, laser-like predictive picture of the election that removes all error from the polls. It's more about giving people a good understanding of how the polls could be wrong and what would happen if they were wrong. By analyzing possible errors and uncertainties in the polls, these models help us approach the election with a clearer sense of how likely each side is to win (and by how much).

As we enter the final week of the election, it's a good time to remember that uncertainty is an inherent part of polls and elections. That's especially true this year as races in swing states are deadlocked. Given that the polls are imperfect, we expect them to vary by some degree in both directions. And if the polls end up being wrong, there's a pretty big margin in the Electoral College results given how close the election is.

In other words, we can summarize the current state of the race like this: Although Trump and Harris have roughly equal chances of winning the election, the final margin will not necessarily be close. In fact, there's a pretty high chance that this won't be the case.

Footnotes

*We simulate potential voting errors for future elections using a fat-tail distribution – specifically, a Student's t distribution with five degrees of freedom (a parameter that increases or decreases the likelihood of surprising tail events in our simulations). This 3.8-point error is the spread or sigma of this distribution – analogous to the standard deviation of a normal distribution. 538's distributions are slightly wider than those used by other forecasting models. This is because our model takes into account the fact that voting errors have become larger over the last decade. Therefore, our model expects more polling errors than if we assumed a constant level of error over time, like most other forecasts.