Will the Polls Be Wrong Again in 2022?

As the saying goes, elections results are the only polls that really matter. Photo: Joseph Prezioso/AFP via Getty Images

Some of Democrats’ recent optimism about the 2022 midterms stemmed from real-life developments, including overperformance by Democratic candidates in special elections and the palpable sense that the issue landscape after Dobbs ended federal abortion rights is significantly more favorable to them. But the belief that we are approaching an atypical midterm in which the president’s party might do pretty well is also based on polls — particularly the generic congressional ballot, an approximation of the national House popular vote. From November 2021 through July 2022, Republicans led in the generic ballot in the RealClearPolitics averages. But now Democrats have a 0.4 percent lead in the RCP averages, and a more substantial 1.3 percent in FiveThirtyEight’s averages. Democrats are also doing quite well in public polling of individual Senate races. Their candidates are leading solidly in Arizona, Pennsylvania, and Wisconsin, and narrowly in Georgia, Nevada, and Ohio; they are also surprisingly competitive in polls of Florida and North Carolina. The purely poll-based version of FiveThirtyEight’s 2022 forecast gives Democrats an 80 percent probability of maintaining control of the Senate, and a 39 percent chance of hanging onto the House.

For context, the polls are telling us that Democrats are doing nearly as well as Hillary Clinton was in the stretch run of the 2016 presidential race; the most wildly optimistic midterms forecasts approach the landslide the polls told us Joe Biden was likely to win in 2020. In other words, a lot of Democratic optimism is based on the assumption that 2022 polls are more accurate than 2016 or 2020 polls. And as Nate Cohn of the New York Times points out, that assumption could be wrong once again:

Early in the 2020 cycle, we noticed that Joe Biden seemed to be outperforming Mrs. Clinton in the same places where the polls overestimated her four years earlier. That pattern didn’t necessarily mean the polls would be wrong — it could have just reflected Mr. Biden’s promised strength among white working-class voters, for instance — but it was a warning sign.

That warning sign is flashing again: Democratic Senate candidates are outrunning expectations in the same places where the polls overestimated Mr. Biden in 2020 and Mrs. Clinton in 2016.

It’s a warning sign, to be sure, particularly in Arizona, Nevada, Ohio, and Pennsylvania, all 2022 Senate battleground states where 2016 and 2020 polling errors in favor of Democrats were pretty egregious. But it remains unclear why these polling errors occurred in the first place. After 2016, the prevailing wisdom was that pollsters were undersampling the white working-class voters where Trump was making big gains over past Republican candidates. Many pollsters changed their methodologies to account for that problem, leaving analysts conflicted and/or baffled by the even larger polling errors of 2020. To be clear, 2020 was a weird year, with the explosion of pandemic-driven voting by mail making traditional polling (and more precisely, estimates of turnout patterns) significantly more difficult. A postmortem by five Democratic campaign pollsters reached no clear conclusion, other than to note that polling errors were most extreme in red states. One entirely plausible theory was that Donald Trump’s endless demonization of media and of all polls in which he was not doing well led many of his supporters to disproportionately refuse to participate in surveys, much as they disproportionately refused to cast mail ballots after Trump’s attack on that method of voting.

But Trump is not on any ballot (directly, at least) in 2022, and it’s worth noting that polling in 2018 — the last midterm when Trump was in office —was quite accurate. In that year, moreover, Democrats overperformed their standing in the polls, winning the national House popular vote by 8.4 percent after leading the generic congressional ballot by 7.3 percent (in the RCP averages). If polls these days systemically underestimate Republican voting strength, that shouldn’t have happened.

The 2020 errors touched off another spasm of know-nothing (but influential) talk about abolishing polls altogether or massively downgrading reliance on them. But the performance of polls since then has been hard to cleanly characterize. Publicly released surveys were far off the mark in the September 2021 California gubernatorial-recall election — but this time, the typical error was in greatly underestimating Democratic turnout. In the fall 2021 off-year statewide elections, polls were quite accurate in Virginia, but not so much in New Jersey (where Republicans lost but overperformed in the polls).

In sum, it’s hard to predict whether the polls will be wrong in November and if so, which way the errors will cut. The contrarian Nate Cohn is clearly trying to throw cold water on Democratic optimism by citing the 2016–2020 polling error that underestimated the Republican vote in certain states that are holding crucial 2022 Senate contests. And he may be right if the incessant and ever-growing distrust among Republicans of all election institutions — from ballot designers to vote counters to media analysts to pollsters — that we’ve seen since November of 2020 means the GOP base will now routinely shirk polls faster than pollsters can adjust for their absence in the samples. The doubts we have about polls reflect broader doubts about the “Trump effect” in 2022. He’s not on the ballot, but he is very present in the political discourse, and even as a malevolent ghost, he may have permanently changed how his supporters think about voting, responding to polls, or (dis-)respecting election results.

As always, it’s smart to pay attention to polling averages rather than individual polls; to take margins of error seriously; and to remember that structural factors (like gerrymandering in House races and the Electoral College in presidential races) mean you cannot mechanically transfer popular votes into results. One thing is certain: The polling industry is praying for a good year in terms of accuracy, lest we consign preelection analysis and postelection interpretation to the terrifyingly arbitrary gods of spin and wish fulfillment.