The Accurate Election Polls That No One Believed

During the past three election cycles, also known as the Trump era, polling developed into something akin to an addiction for political junkies and concerned citizens alike. But its accuracy appeared to be declining, in ways that have caused acute psychological harm to many Democrats—each of the previous three national elections featured polling averages that, to varying degrees, understated Republican support. In the 2016 Presidential election, Donald Trump shocked Hillary Clinton by outperforming the polls in a number of states, especially in the Midwest and Pennsylvania, with their high concentration of white voters without college degrees. In 2020, Trump lost to Joe Biden, but beat his national polls by more than three points, and did even better than that in many swing states—again in the Midwest, and in Florida. In between, in the 2018 midterms, polls were more accurate but missed Republican strength in several critical battlegrounds; Democrats failed to recapture the Senate that year, even while picking up forty-one seats in the House. (Polls also underestimated Republicans in 2014, in the pre-Trump midterms.) This week, the mystery was not simply whether the polls would come within respectable proximity to the result; it was whether they would continue to be off in the same direction, and further disappoint Democrats already bracing for a “red wave.”

First, a caveat: many of the big, brand-name pollsters conducted considerably fewer polls this cycle. According to FiveThirtyEight, in the 2010 midterms, legacy media sponsored eighty-five Senate polls between May and late October. This year? Twenty-five. (There are probably a number of reasons why, not least because telephone polling, with its decreasing response rates, is getting more expensive.) Throughout the past twelve years, FiveThirtyEight estimated that the share of Senate polls “conducted or sponsored by campaigns or partisan organizations” has grown by more than three hundred per cent. This means that many of this year’s polling averages comprised work done by firms not associated with traditional media outlets or educational institutions, often with ties to Republicans—and likely skewed in that direction. “It certainly felt like there were more Republican polls out there,” Jeff Horwitt, who is a Democratic pollster for NBC News, and a partner at Hart Research Associates, told me.

Although the averages consisted of fewer polls, and fewer high-quality polls, they appear to have been relatively accurate. In fact, they were perhaps slightly too favorable to Republicans, even in states where there has recently been significant polling error understating the G.O.P. vote. In Ohio, J. D. Vance beat the Democrat Tim Ryan by around six and a half points. According to FiveThirtyEight, the final polling average in the state had Vance winning by 6.2 points. (In 2020, Trump outperformed the final polls in Ohio by seven points.) In Wisconsin, where Trump beat his final polling averages by about six points, this year’s polls showed Senator Ron Johnson, the Republican incumbent, leading by around three and a half points. He won by one point. (Tony Evers, the Democratic governor who was just reëlected, ran ahead of his final polling average by several points.)

In North Carolina, a state with some history of polling error favoring Democrats in 2016 and 2020, the Republican Ted Budd won by around four points—approximately what the polls were predicting. In Georgia, Raphael Warnock and Herschel Walker are headed to a runoff after ending up nearly tied—again, very close to the polling averages. (In the governor’s race there, the Republican Brian Kemp easily defeated the Democrat Stacey Abrams by a margin that also tracked closely with the polls.) We don’t yet have final results in Nevada, but its race for the Senate is razor-thin, as the polling suggested.

Some states were off by a greater amount—but not always in predictable ways. Polls in Pennsylvania have overstated Democratic support in the past two Presidential elections. On Tuesday, the final polling averages showed a half point lead for the Republican Mehmet Oz. But John Fetterman, the Democrat, is on track to win by four. Similarly, in New Hampshire, the Democrat Maggie Hassan is headed to a nine-point victory—beating her final polling average by seven points. It’s not clear exactly why these results differed from the poll averages. During the weekend, Nate Cohn, the analyst who runs the Times’ polling operation, noted that in the last days of the Pennsylvania race, four conventional pollsters released surveys showing Fetterman tied or leading, whereas lesser-known, Republican-leaning outlets and firms had Oz ahead. This discrepancy suggests that the polling averages might have been even more accurate if the traditional firms had conducted more surveys, given that the ones many of them did release were more favorable to Democrats.

There was at least one crucial state where Republicans markedly outran their poll numbers: Florida. In the Senate race there, FiveThirtyEight had Marco Rubio leading Val Demings by nearly nine points. In the governor’s race, Ron DeSantis was ahead by twelve. Rubio won by seventeen points, and DeSantis by almost twenty. With few big media organizations polling the state, you’d expect the smaller, Republican-tilting surveys to overstate Republican support. And still Republicans exceeded expectations.

What does this all add up to? This cycle’s polls were certainly more reliable in swing states than they were in 2016 and 2020, and this was especially true of some of the nonpartisan firms that have recently struggled, such as Marist and the Times/Siena College. But the results across the country don’t make it easy to draw conclusions about why, on balance, the polls fared better this time. Because so few nonpartisan polls were taken, there may not be enough good data to meaningfully match against the final results. In other words, we don’t know why the nonpartisan polls were more accurate. “Low response rates are still things that are there,” Horwitt told me. “I don’t think polling is ‘fixed’ now.” But, he added, “I don’t think it was totally broken two years ago or four years ago.”

What is clear is that the polls, such as they were, provided a decent view of what was happening in most of the Senate and gubernatorial races. (The House numbers are still incomplete, but the national outcome also seems on track with what the polls predicted.) A week ago, if you had been looking solely at polls and given less thought to the general political environment—a shaky economy, an angry populace, an unpopular President leading his party through his first midterm—you would have come away with a better picture of what happened on Tuesday. It’s understandable that so many observers, especially liberals, had doubts about the polls, which is probably why the results came as such a shock. Whether the polls being more accurate than the “vibes” is comforting or scary depends on how you felt about polls to begin with. ♦