4. What’s the deal with election polls?

An election poll is pretty much what it sounds like: a poll conducted before (or sometimes after) an election that focuses on the election. You’ll often hear about election polls when they’re being cited to show who’s ahead and who’s behind in the race – the “horse race.”

Though it often attracts the most attention, discovering the level of support each candidate has – or even forecasting the outcome of the election – isn’t the most important reason election polls are conducted.?

What are election polls used for?

Election polls are particularly useful for understanding the meaning of the election. Polls can help clarify what voters – and nonvoters – are saying with their actions.??

Here’s a short list of how election polls are used:?

• They help us identify what kinds of people are voting for which candidate.??

• They help explain the issues and concerns that may be leading to a choice for a particular candidate.??

• They provide insight into why some people vote and others do not, including differences in engagement and interest between supporters of each candidate.?

• They reveal how, when and where people cast their ballots – for example, whether it’s on Election Day or early, or by mail or in person.?

• They describe how satisfied people are with their choice of candidates, for instance, or whether they find a campaign informative.?

• They can tell us how the public judges the election process and reacts to the outcome of the election.?

• For those involved in election campaigns, polls can also provide important strategic information: where and when to campaign, which groups to try to appeal to, which messages are likely to work.?

How are election polls conducted??

The quick answer is … just like other polls.

Like all polls, election polls have to gather an accurate sample of the population and an accurate measurement of the subjects of interest. To achieve this accuracy, election polling has evolved just like other polling has.

We tracked the methods used by polling organizations in election years between 2000 and 2022 and found that online election polling – especially with opt-in sampling – grew rapidly after 2008 and is now widespread. Live interviewer telephone polling, on the other hand, dwindled after 2012.?

Beyond the growth of online opt-in polling, a couple of other things are apparent in this colorful chart. One is growth in the sheer number of pollsters conducting national election polls – more than twice as many in 2022 as in 2000. Cheap opt-in methods have allowed a lot more people to claim the title of “pollster.” The other thing that stands out is growth in the diversity of methods now in use. Pollsters are trying out lots of approaches!?

What’s different about election polls??

There is one thing that’s unique about preelection polls. Because most of them purport to describe the views of the people who will actually vote, rather than the general public, they have to determine which people among those interviewed will actually turn out. This is not an easy task.?

Even with voter turnout in 2020 that eclipsed 100-year records, a third of the U.S. eligible adult population did not cast a ballot. Not everyone who says in a poll that they intend to vote actually does, and some people who tell a pollster they don’t plan to vote ultimately do.

Furthermore, as a group, nonvoters and irregular voters often differ from regular voters in terms of which candidate they prefer. Figuring out who will really vote can make a difference in what the poll finds about, say, the composition of the voting public. That may be a particularly hard job in a high-turnout election, when people who haven’t voted in the recent past suddenly show up, confounding pollster estimates.?

Are election polls accurate??

You might be surprised to hear that election polling in the U.S. and around the world has a good track record for accuracy. A comprehensive study found that polls taken shortly before an election had an average error of less than 2 percentage points in estimates of support for a given party or candidate. But preelection polling was less accurate in the 2016 and 2020 presidential elections than it had been in the previous several contests, overstating support for Democratic candidates. Historically, polling errors have sometimes favored Democrats and sometimes Republicans, with no clear tendency.??

The errors in 2016 mostly affected state polls, but many of the errors were in critical battleground states that contributed to Donald Trump winning the presidency. National polling that year was quite accurate, but quantitative forecasts of the election – some of which gave Hillary Clinton a greater than 90% chance of winning – may have undermined public confidence in the polls by leading many observers to believe the outcome was far more certain than the polls actually indicated.??

Despite efforts to correct the problems of 2016, even larger errors affected both the state and national polls in 2020.?

Pollsters scrambled to improve their methods and in 2022 produced much more accurate results in key state elections. But fears remain that presidential polling may again struggle to adequately represent support for Trump.?

Because – fairly or unfairly – the outcome of elections is used to judge the accuracy of polling, it’s useful to understand the size of the errors we are talking about. According to the American Association for Public Opinion Research (or AAPOR, the professional association for the survey research field), national polls in 2020, on average, overstated Joe Biden’s margin over Trump by 3.9 percentage points, the largest such error since 1980. Even though his support was overstated in the polls, Biden won the election, so the polling errors were not as obvious as in 2016. But with a public closely divided between the two political parties, it’s especially crucial for the polling community to minimize the size of the errors.?

At the same time, it’s important for polling consumers (that’s you!) to understand polling’s limitations. The margin of error you may have heard about is often as big as or bigger than the difference in support between two candidates in a close race. And the traditional margin of error doesn’t capture all of the sources of error affecting polls. ?

Extra credit?

Want to learn even more? Here’s a reading list:?

• The Death Of Polling Is Greatly Exaggerated?(FiveThirtyEight)?
• Can Likely Voter Models Be Improved??(Pew Research Center)?
• Q&A: A conversation about U.S. election polling problems in 2020?(Pew Research Center)?
• What 2020’s Election Poll Errors Tell Us About the Accuracy of Issue Polling?(Pew Research Center)?