Skip to main content
Menu
You Can Pay Attention To Those Trump vs. Biden Polls. But Be Cautious.

We are now about six months from the general election,1 and while the national party conventions won’t make it official until August, President Trump will face former Vice President Joe Biden in November (barring something very unexpected happening). So the question we’re left with now is: When should you start paying attention to the general election polls?

We at FiveThirtyEight have cautioned you not to take early general election polls too seriously, but the answer for when you should tune in isn’t exactly straightforward either. There’s actually a pretty big debate over just how meaningful early general election polls are. MIT researcher Alexander Agadjanian and The Economist data journalist G. Elliott Morris argue that early polls in recent cycles have been closer to the final outcome than polls from previous cycles and should therefore be taken seriously. But political scientists Robert Erikson and Christopher Wlezien maintain that while early general election polls might be more accurate now, they’re still not as accurate as those conducted closer to Election Day, and that distinction matters as even small shifts in the polls can matter a great deal to the final outcome in our era of highly competitive elections.

So to better understand just how meaningful early general election polling is we did our own analysis, collecting all the national surveys we could find from 1980 to 2016, spanning from 200 days before the presidential election to the day before. We then created a rolling, seven-day polling average of the margin between the Democratic and Republican nominees2 as that allowed us to calculate just how far off the polls were from the final Election Day vote share margin on any given day.3 And what we found is, well … each side in this debate has a point. A satisfying answer, we know.

On the one hand, the margin in early general elections polls has been closer to the final national popular vote margin in recent cycles since at least 2004 — you can see this quite clearly in the chart below.

There just isn’t as much movement in the polling margin in recent elections. But even when early polls have been closer to the final outcome, the error has still often been large enough to swing an election from one party to the other. What’s more, the direction of the polling error has been inconsistent — sometimes polls tend to be Republican-leaning; at other times, Democratic-leaning. This error is substantial enough that it makes it hard to predict the final election result, even if the early polls aren’t that far off.

Take the last three presidential elections. In 2016, the polls about six months from the general election were about 2 points more Democratic than the final national popular vote margin (2.1 points), on average, so while they weren’t far off, the size of the error was about the same as the final margin in what was a very close race. But in 2008 and 2012, the early polls missed in a Republican direction: In 2008, Barack Obama won by 7.3 points, but then the average six months out was about 6 points more Republican than the final outcome. Similarly, in 2012, Obama won by 3.9 points, but the early polls were, on average, 2 points more Republican compared to the final margin.

The difference at the margins, in other words, can still be very meaningful. Consider the 2016 election. Trump lost the popular vote by 2 points but still won the presidency because our elections are determined by the Electoral College, not the national popular vote. It’s harder to imagine this being the case, though, had Trump lost the popular vote by 4 points, which was Clinton’s average lead between 150 and 200 days out.

Why did Clinton’s edge change from 4 points in the early surveys to 2 points in the final result? Well, for a variety of reasons. Polls are snapshots in time, while campaigns are dynamic events. And although some moments are somewhat predictable in their impact — such as the convention bounce — many others aren’t. This seems particularly important right now because we don’t know how the coronavirus pandemic will affect the standing of the two major-party nominees. And unexpected events can’t be discounted: Clinton may have lost in 2016 in part because of the Comey Letter, which came out less than two weeks before the election.

All of this is to say it’s difficult to know how 2020 will shape up. Some pollsters have adjusted their methodologies after 2016 — such as weighting samples by educational attainment — and the 2020 race will happen under a different set of circumstances, so it’s not a given that polls will lean Democratic as they did four years ago. But what the polls can tell us right now is that the 2020 race should be pretty competitive. As the campaign develops over the next few months, the electorate’s preferences could shift in small but meaningful ways that could ultimately be consequential. So look at the early polls — you know you’re going to anyway — but know that they are subject to change.


Should we sacrifice our data privacy to fight the coronavirus?

Footnotes

  1. 190 days, to be exact.

  2. Specifically, we used the end date of a survey in our averages. If a pollster had more than one poll in a seven-day period, we included only the most recent survey. We also used the most inclusive polls possible, provided all candidates listed ran in the general election.

  3. That is, we calculated the difference between the margin in the polling average and the final Election Day vote share margin, not the sampling error inherent to all polling, though that is something else to keep in mind when looking at poll movement.

Geoffrey Skelley is an elections analyst at FiveThirtyEight.

Comments