How wrong can opinion polls be in America?

DOpinion polls Given that gamblers are essentially equal, gamblers who bet on the results of the US presidential election have become increasingly confident that Donald Trump, the Republican candidate, will win. PolyMarket, the prediction market that saw more than $2.6 billion traded in the election, gives it a two-in-three chance. In fact, gamblers are betting that the polls are underestimating him for the third time in a row.

Such an error is certainly possible. Polling averages show either Kamala Harris or Trump leading in each of the seven swing states by margins smaller than normal polling error. Democrats fear a repeat of the dramatic failures in the polls in 2016 and 2020, when Trump performed better than expected. But there is no guarantee that the error will be in the same direction this year: pollsters have gone to great lengths to account for past errors.

Opinion polls are conducted by polling a representative sample of voters. Errors can arise in several ways. There is normal statistical variation, which affects all polls, especially those with small sample sizes. There is a risk of last-minute fluctuations or unexpected demand patterns. The biggest problem facing pollsters is ensuring that their samples are representative. Researchers are doing what they can to do just that: finding new ways to reach voters from certain demographic groups and using weighted sampling to increase the relative importance of underrepresented groups.

FiveThirtyEight, a data journalism organization, has calculated polling averages for presidential elections dating back to 1976 (see chart). On average, the gap between exit poll results and the actual margin of victory is 2.7 percentage points nationwide and 4.2 points in individual states. Now, FiveThirtyEight estimates that the largest lead for either candidate in the seven swing states is just 1.5 points, for Trump in Arizona.

Polls in 2016 and 2020 systematically underestimated Trump’s vote, especially in battleground states. After the 2016 elections, an autopsy was conducted by AborA polling organization noted a late swing toward the Republican candidate and an overrepresentation of graduates in poll samples. Most companies have begun weighting their samples to do a better job of reflecting the educational level of voters.

In 2020, Trump was repeatedly underestimated, but for different reasons. this time Abor Non-response bias was identified – Republican voters were less likely to respond to polls. One theory is that they were less likely than Biden voters to be home during the COVID-19 pandemic (twiddling their thumbs and responding to polls). Another reason is that Republican voters do not trust pollsters, which discourages them from answering surveys.

Since 2020, poll organizers have made every effort to reach a representative sample. They experimented with recruitment that appealed to specific sectors of society (postcards filled with patriotic images, for example) and new methods, such as text messaging. Whether this will be enough to account for Democratic bias in response rates or whether Trump supporters will still be reluctant to answer polls is anyone’s guess. If the mistakes seen in 2020 or 2016 are repeated to even a small degree, it would be disastrous for Ms. Harris.

There are also plenty of plausible scenarios in which polls underestimate Harris’ support. For example, errors that occurred in 2020 may have been specifically related to the pandemic. Pollsters may have overcorrected it since then. Opinion polls can be frustrating, but without them we would not be able to say with such confidence that the election outcome is likely. ■

Stay up to date on US politics with The we In short, our daily newsletter with quick analysis of the most important election stories, Checks and Balance, a weekly note from our Lexington columnist on the state of American democracy and the issues that matter to voters.