President Biden’s re-election campaign is in jeopardy because of his poor debate performance. But lurking behind the Democratic anxiety is another factor: the polls.
The average of polls has shown Donald J. Trump with an almost uninterrupted lead for 10 months, and this undoubtedly shapes the calculation of Democrats in Washington. It’s hard to believe there would be any serious drive to replace Mr. Biden if the polls showed him clearly leading, as they did throughout the last election.
This puts a big responsibility on polls and pollsters. The fate of the president may not rest in their hands, but — at the very least — it will be influenced by their methods. In a sense, it is what polling is supposed to do in a democracy: give political actors insight into the will of the people, enabling decision makers to better respond to the demands of the electorate.
At the same time, it’s not obvious that polling is up to the challenge.
Polls are inherently imprecise measurements, subject to countless sources of uncertainty. While they continue to be the best way — or only way — to measure public opinion in a big, diverse country, polls have been wrong before, and they will be wrong again.
And at this stage, Mr. Biden’s chances increasingly rely on the possibility that the polls are simply “wrong.” He could retake the lead in the polling before November, of course, but given the stability in the polls so far, it may be unrealistic to expect such a pronounced shift toward him.
Mr. Biden himself appears to be counting on the polls being “wrong.” In his interview with George Stephanopoulos of ABC last week, he questioned whether polling was “as accurate as it used to be.”
Is he right? Will the polls be wrong again?
The unsatisfying truth is that no one knows.
It’s extremely difficult to anticipate polling errors in advance. We won’t know whether this is a “good” or “bad” year for the polls until the final results in November. That said, the polls are usually more right than wrong, even in their “bad” years.
Here’s what we do know about the state of the polling — and the likelihood of a polling misfire — heading into what could be a pivotal stretch in American history.
Who’s ahead in the polls?
Mr. Trump’s longstanding lead has grown since the debate. Polling has been sparse since then, as the Fourth of July holiday weekend put a pause on most high-quality polling. But Mr. Trump leads by about three percentage points in the national polls taken over the last 10 days. He’s ahead by even more in most of the battleground states likely to decide the Electoral College.
More ominously for Mr. Biden, Mr. Trump is faring even better in the highest-rated polls. That includes the New York Times/Siena College poll, which showed Mr. Trump up by six points among likely voters, and CNN/SSRS and Wall Street Journal polls, which showed Mr. Trump up six among registered voters. A bipartisan poll in Wisconsin — fielded partly by a Biden pollster — found Mr. Trump up six points there as well (when including minor-party candidates). The highly regarded Selzer poll found Mr. Trump up by 18 points in Iowa, even before the debate.
What about the private polls?
The public polls you see every day are a sliver of the data that’s out there. The campaigns, campaign committees and outside groups all sponsor their own national, state and district polls, which gradually if incompletely percolate out to other political actors.
These private polls have just as much influence on Mr. Biden’s standing as the public polls. They might even have more, especially for the members of the Senate and House, who naturally trust their own polls or those that they hear about from their colleagues.
What do the private polls show? I’m generally reluctant to write about anything I hear about in the world of private polling. I talk to pollsters all the time, but I hear only fragments of the broader picture, and it’s impossible to know whether what I’m hearing is representative.
That said, I see zero indication that the private polling is materially different from the public picture — something I’ve heard all cycle. Some of what I hear about is a little better for Mr. Biden, some of it is a little worse, and some of it is much worse — especially in New York and California, like a poll in the California congressman Mark Takano’s district showing Mr. Trump leading in a place Mr. Biden won in 2020 by 26 points.
Of the three buckets, I’ve heard the most lately about polls in the last category of “much worse,” but this may reflect that people are most eager to share and gossip about the most eye-popping numbers.
How accurate are the polls?
It’s worth dividing this question into two parts.
First: How accurate are the polls at this stage, four months from the election?
Historically, the short answer is that they’re hardly indicative of the final margin.
The longer answer must add an important caveat: Early polls have been getting more accurate. Over the last three cycles, early polls have even been about as accurate as the polls taken immediately before the election. This reflects a higher level of political engagement, better-known candidates and growing partisan polarization. This cycle, the extraordinary stability in the early polls suggests this trend will continue.
Second: How accurate are the final polls?
The short answer: They’re pretty accurate. Since 1972, the final national polling average in a presidential election differs from the national vote margin by an average of about 2.5 points.
The longer answer, however, must note that 2.5 points is extremely consequential in our era of relatively close elections. Four of the last six presidential elections have been decided by less than 2.5 points in the key battleground states. And in recent cycles, polls in the critical Northern battleground states, like Wisconsin, have been even less accurate than the national polls.
Are the polls really less accurate than they used to be?
It’s not as clear as you might think.
As I mentioned, the final national polling averages have been off by about 2.5 points in modern elections. By that measure, the polls have been slightly below average in accuracy in recent presidential elections, but not by much: In 2020, they were off by four points; in 2016, two points; in 2012, three points. None of these tallies are especially unusual historically — 1980 featured an eight-point polling error.
It’s conceivable polling is much worse than it used to be, even if elections are infrequent enough that the scope of decline isn’t evident. If the death of polling really is at hand, it would be straightforward to explain. Over the last 15 years, the response rates to telephone surveys have plummeted, and the survey research industry hasn’t regained its footing. High-quality research has become more expensive, and polling averages have been flooded by cheap, low-quality internet polls. The high-quality polls, meanwhile, rely more on statistical adjustments to ensure a representative sample, rather than the principles of random sampling that underpinned survey research in the first place.
We may look back on 2024 as the year that put the nail in the coffin of political polling, but it hasn’t happened yet. And two years ago, the 2022 midterm elections featured some of the most accurate polling on record, especially from high-quality nonpartisan pollsters.
Did the polls fix what went wrong in 2020?
The short answer is no.
It may not be clear whether the polls are getting worse, but this much is clear: The polls in 2020 were terrible.
National polls overestimated Mr. Biden’s winning margin by around four points. The error was even larger in the key battlegrounds. All considered, it was the worst year for the polls in a competitive presidential race since 1980.
Worst of all: It’s not clear why the polls were wrong. As the postmortem of professional pollsters put it: “Identifying conclusively why polls overstated the Democratic-Republican margin relative to the certified vote appears to be impossible.”
Without any solid explanation for the polling error, pollsters have blamed the misfire on something called nonresponse bias — basically, the idea that Mr. Trump’s supporters were significantly less likely to respond to surveys than demographically similar Biden voters. This is a diagnosis of exclusion: If other explanations — turnout, undecided voters, demographic weighting, and so on — can be ruled out, there must have been some unknown response pattern biasing the polls.
The source of nonresponse bias in 2020 is hotly debated, and depending on the explanation it may or may not pose a challenge in 2024. One popular theory blames the pandemic, as Democrats were disproportionately likely to stay at home — and take polls. If so, polling would return to normal with the end of the pandemic. Another theory says Mr. Trump’s supporters would simply never trust the media or pollsters enough to take a survey. If so, polls might still be underestimating him, even now. There are plenty of other hypotheses in between.
Are there signs of nonresponse bias again?
Not necessarily, but it’s hard to know.
Pollsters may not have fixed what went wrong in 2020, but they are taking the challenge of nonresponse seriously. As a result, there are a few signs that polling may be on safer ground right now.
One indication: response rates by party. Looking back at Times/Siena data from 2020, response rates were about 20 percent higher among Democrats than Republicans. We account for Democrats or Republicans being likelier to respond, of course, but in hindsight it was the only indicator in the data that was a possible warning sign that the polls could be off. So far this cycle, Democrats and Republicans have responded to surveys at nearly equal rates.
Another indication: the growing number of high-incentive surveys that obtain higher response rates than the usual polls, like the Times/Ipsos survey of Wisconsin in 2022 that paid voters up to $25. These surveys ought to be less vulnerable to nonresponse bias, and if they differ from typical surveys, it could be a warning sign. So far, they have offered similar results.
The latest example: the Pew NPORS study, out earlier this week. It obtained a 30 percent response rate by paying a cash incentive both up front and when the survey was completed. It found Republicans with a one-point advantage in party identification among all adults and a two-point edge among registered voters — a tally in line with Times/Siena polls over the last year.
What about the turnout?
Nonresponse isn’t the only reason polls can be off. One obvious example is turnout: In theory, polls can take a perfect sample of the full population but still be off-target if the makeup of the electorate is different from what the pollster anticipated.
In recent cycles, turnout hasn’t been a big factor in polling error — at least according to our data. This year, there’s reason to think it could be a much bigger challenge.
That’s because Mr. Trump’s lead is built on strength among relatively disengaged, lower-turnout voters. Even after the debate, Mr. Biden leads among voters who turned out in the midterm election, according to Times/Siena data. Mr. Trump counters with a nearly 20-point lead among those who did not vote in the midterms.
Mr. Trump’s wide lead in the Times/Siena poll thus depends on whether these non-midterm Trump supporters voters show up in sufficient numbers. Many undoubtedly will: Mr. Trump led by six points in the poll among the broader group of voters who turned out in 2020, and he led by seven points among registered voters who said they were “almost certain” to vote.
Even so, the details will really matter. The exact number and character of the turnout among irregular voters could greatly affect the outcome in 2024, given the huge difference in preference between regular and irregular voters.
What about the undecided voters?
Undecided voters are another source of potential polling error. In theory, a pollster can have a perfect understanding of who will vote but still appear “wrong” on Election Day if voters change their minds or if undecided voters break one way or another.
Here again, there’s reason to be more concerned than in recent cycles. There’s a large number of voters who dislike both candidates — the so-called double haters — and their preferences could easily prove to be unsteady. There’s also the third-party factor: Many voters who dislike the major-party candidates may agonize over whether to back Robert F. Kennedy Jr. or another minor-party candidate. And at this early stage, many voters aren’t tuned into the race.
Although the race has been notably stable to this point, the large number of double haters creates the conditions for volatility. Mr. Trump’s strength among young, Black and Hispanic voters, in particular, stands out as a possible vulnerability. These voters have traditionally voted for Democrats, they hail from Democratic constituencies, and in many cases they continue to identify as Democrats or Democratic leaners. They say they’re voting for Mr. Trump — but will they really?
We won’t know until election night. We never do.
The post Polls May Shape Biden’s Future. Are They Up to the Job? appeared first on New York Times.
