There’s a decent case that the polls were the best they’ve ever been in 2024.
That’s not necessarily true by the usual measures. When judged against the final results, the polls missed by a more or less average margin. And, worryingly, they systematically underestimated support for Donald J. Trump for the third cycle in a row.
But by a less conventional measure — value more than accuracy — the polls excelled. They were fundamentally right about a surprising and important story about the election — a story we might not have imagined, let alone expected, without them.
Let’s go through a quick list:
-
Big Trump gains among young and nonwhite voters and a huge decline in racial polarization? Check. In the end, Mr. Trump surged among Hispanic and Asian voters; turnout plunged among Black voters; and the partisan gap between white and nonwhite voters fell to its lowest level since the Civil Rights Act in 1964.
-
A Trump advantage among low-turnout voters, one so great that it merited disregarding Democratic strength in special elections and off-year general elections? Check.
-
Surprising gains for Mr. Trump in the places where Republicans excelled in the midterm election, like Florida and New York City? Check.
-
A reduced gap between the popular vote and the Electoral College, and even a Trump popular vote victory? Check. In the end, Mr. Trump’s advantage in the Electoral College almost evaporated. There was only a small gap between the “tipping-point” state — Pennsylvania, which Mr. Trump won by 1.7 points — and the national popular vote, which Mr. Trump leads by 1.5 points.
-
A national political environment that was deeply unfavorable to Democrats, including a Republican advantage in party identification? Check. Whether by the exit polls, the AP/VoteCast data or the pre-election polls, Republicans now outnumber Democrats in America for the first time in decades.
Maybe you, dear reader, would have seen all of this coming without the polls. As for me? Not a chance.
Two years ago, I thought a Biden-Trump rematch would probably look a lot like the 2020 election. It wouldn’t be an exact rerun, of course, but I would have guessed it would be pretty similar.
From the start, the polls said a Biden-Trump rematch would be nothing like the 2020 election. Every single one of these surprising shifts was clear in the polls more than a year ahead of the election. It’s hard to think of a cycle when so many polls, so far ahead of time, wound up telling the tale.
Importantly, many alternative measures didn’t tell the tale. Fund-raising, crowds, special elections and the Washington primary all augured well for Democrats. The assumption the electorate would simply behave as it did in the past also boded well for Democrats.
We’ll get to where the polls fell short, but in some basic sense this is what polling is all about: It’s a way to learn something about the country and public opinion that we wouldn’t have known if we had talked only to our friends or neighbors, or stuck to our partisan priors.
By this measure, I can’t think of an election where polling did better.
OK, now about that miss
There’s another measure by which the polls didn’t necessarily excel: The polls underestimated Mr. Trump yet again.
On average, the polls underestimated him by about two percentage points or so. Our Times/Siena polls were actually a bit worse by this measure, underestimating him by about 2.9 points.
It’s worth noting that this isn’t necessarily a bad performance for the polls. It’s basically the average error historically, which is better than one could say for 2016 or 2020. The polls were actually about as good as they were in 2012, which was regarded as a major success for the industry. Still, even if it is a fairly typical error, it takes on greater significance as the third consecutive election in which polls underestimated Mr. Trump.
What happened?
Before the election, we outlined the two basic theories of polling error in 2016 and 2020. One, which I called the “unified theory,” essentially said the error was because Trump voters are simply much harder to reach than other kinds of voters — and even than other kinds of Republicans. The other, which I called the “patchwork theory,” said the misses in 2020 were much different from the misses in 2016, and could mostly be attributed to the pandemic.
In the final account, my guess is that both theories will show their merit.
The patchwork theory helps explain why the error was quite a bit smaller than four or eight years ago. Under the most extreme version of the unified theory of hard-to-reach Trump voters, it’s difficult to explain how the polls were reasonably accurate — off by two points or so — in states where they erred badly in the prior two presidential elections, like Wisconsin or Pennsylvania.
Many pollsters also probably benefited from weighting on recalled vote — a technique that tends to yield stronger poll results for the party that lost the last election (in this case, the Republicans). Since polls underestimated Mr. Trump yet again, it’s likely that many pollsters who gambled on the measure found it paid off — even as it came with the cost of obscuring major shifts in states like Florida, Texas or New York. (In Times/Siena polling, it would have made little difference on average — helping some polls, hurting others, but often making little difference.)
At the same time, the theory of hard-to-reach Trump voters offers a compelling explanation for yet another underestimation of Mr. Trump. As long as his support is so closely correlated with political engagement, the polls will struggle to fully capture his support — at least as long as those voters show up.
On that last point, there is one additional variable to consider: turnout. The lower turnout in many Democratic areas raises the possibility that the electorate was more favorable to Mr. Trump than pre-election polls suggested. If so, it’s possible that pollsters’ estimates of who would actually vote — not merely persistent nonresponse bias like in 2016 or 2020 — might be at least partly responsible for the error.
As we wrote ahead of the election, our Times/Siena polls suggested that “Mr. Trump could win the presidency, perhaps even fairly handily, if he could turn out all registered voters.” Indeed, he led our final Times/Siena polls of Georgia, Michigan, Pennsylvania, Arizona and the national popular vote among registered voters. Kamala Harris, however, fared better among likely voters in the polls: Voters who backed her were likelier to say they would vote and likelier to have a robust track record of voting.
It will be many months until we have a full account of the turnout, but if it ultimately emerges that Mr. Trump benefited from better turnout than expected, this could explain some of the difference between the polls and the final results.
A glass half full
In our era of close elections, it’s hard for pre-election polls to offer a clear answer about which party is going to win.
I know it’s what many of you want. It may be the reason you subscribed to this newsletter! But the polls just aren’t precise enough to do the job in a close race.
Imagine, for instance, that the polls had been perfect — if they had shown Mr. Trump up by 1.7 points in Pennsylvania, 1.5 points nationally, 1.4 points in Michigan, 0.9 points in Wisconsin and so on. Would the polls have really added much additional value? Either way, the election was very close and competitive, and we still would have entered Election Day deeply uncertain about which candidate would ultimately win.
Obviously, nailing the result is ideal; we spend plenty of time trying to figure out how to do so. When the final turnout data arrives, we will undoubtedly return to the question again. Yet in this particular election cycle, the polls could not have accrued much additional value by hitting the result precisely. The race was too close, either way.
With that in mind, there’s a good case to take a glass-half-full perspective on this year’s polling. The polls taught us a lot about what was happening in American politics. They nailed the major demographic and political trends, and they did so more than a year before the election. We wouldn’t have known much about it otherwise.
The post 2024 Review: A Few Misses, but the Polls Got the Big Stories Right appeared first on New York Times.
