Midterm election polls underestimated Democrats by about four percentage points in key contests this November, a break from their fairly accurate performance in the 2018 and 2022 midterms and from the last three presidential elections, in which they underestimated President Donald Trump’s support.
Diagnosing the causes of polling misses often takes many months. But The Pollwatcher asked pollsters from a range of backgrounds for their initial thoughts on why most surveys underestimated Democrats as well as the overall performance of election polls this year.
Want this in your inbox?
Sign up for The Pollwatcher newsletter here.
Likely voter models struggled to quantify Democrats’ turnout advantage
YouGov chief scientist Doug Rivers highlighted the wide range of pre-election estimates that depend on different methods of identifying likely voters.
Likely voter models appear to have had outsized effects in 2026. We published a piece describing the impact of various choices on the size of the Democratic lead. … We were off by about 2 percent on the Democratic vote (D+9 vs. D+13 actual). We think that we were about right on net switching (aka crossover voting — about 4 percent, on vote share), but overestimated Republican turnout and underestimated minority and younger voter turnout.
Emerson College pollster Spencer Kimball also highlighted challenges measuring turnout interest:
Turnout is a moving target; enthusiasm expressed in polls doesn’t necessarily translate to turnout on Election Day. … Record Hispanic turnout exceeded our expectations, but most of the polling caught the Hispanic swing for Democrats by almost 2:1.
Marist Poll director Barbara Carvalho highlighted that the blue wave narrative was rooted in polls:
The blue wave narrative was an early and consistent theme this fall with polls simply differing on the magnitude of Democratic support. That difference was likely based on the degree to which methods included previous turnout estimates … less dependence on previous elections, wider margins.
BlueLabs vice president of polling Joy Wilke highlighted the role of irregular voters:
Infrequent voters have swung dramatically since 2024; in ’24 they broke for Trump by 7-8 points, but in ’26, they broke for the Democrats by 16 pts. That’s a huge shift! And while these infrequent voters are decisive for elections, they’re also really hard to reach!
Atlas Intel CEO Andrei Roman criticized some pollsters for not recognizing a “wave” election:
AtlasIntel polls were balanced in terms of underestimating or overestimating Democrats and Republicans. … Thinking about polls more broadly and not just AtlasIntel polls, most pollsters did indeed massively underestimate Democrats because they missed the basic fact that this was a wave election. This was probably due to poor modeling of turnout by partisanship and race (Democrats and Latinos mobilized much above the average of other groups) but also by age (young voter mobilization) and the swing in the independent vote.
The full picture of turnout will become clear in the coming months as states update their rolls to identify whether every single registered voter turned out. By matching poll respondents to these files, pollsters will be able to assess whether different likely voter models would have better captured the 2026 electorate.
‘Recalled vote’ weighting ranged widely
Weighting poll samples to reflect the results of the most recent presidential election was credited with reducing error in 2024 election polling, but pollsters used widely varying approaches to doing so this fall.
One approach used by Post-Schar School polls and Marist polls is to ask respondents how they voted in 2024 and weight the sample of registered voters to match 2024 turnout and vote choice.
YouGov’s panel-based polls relied on reported vote choice closer to the 2024 election and combined this with a statistical model of the overall population of registered voters. New York Times-Siena College polls use “synthetic past vote,” which combines self-reported support for Trump and Kamala Harris with statistically modeled estimate of support among other 2024 voters. And Fox News’s polling team “monitored 2024 vote recall and when it went beyond a certain range they partially weighted back to 2024 results,” according to a spokesperson.
Each of the above polls weighted samples of registered voters to match 2024 results, then used turnout indicators to produce likely voter estimates for 2026.
Kimball said Emerson College polls “weighted the sample to general voter targets including a split electorate in the national 2024 popular vote, then [applied] a LV model that allows non-2024 voters to fall naturally.” Their final national poll this year had Harris voters outnumbering Trump voters by two points among likely voters.
AtlasIntel weighted likely voter estimates to targets for the share of Trump and Harris voters it expected in 2026. In an email to The Post, CEO Andrei Roman said “the in-depth modeling of turnout is AtlasIntel intellectual property.”
Did recalled vote weighting work in 2026?
Pollsters gave mixed answers to this question, although some said they are still analyzing the results. Survey160 pollsters Kevin Collins and Alejandro Frydman found that weighting registered-voter samples by recalled vote increased accuracy dramatically in three of four states where the firm conducted late polls, while slightly reducing accuracy in a fourth state. AtlasIntel’s Roman said “It helped. It always helps.”
Barbara Carvalho of Marist said the group is still evaluating the impact of the 2024 vote, but her “first glance is it helped in some states and not in others.” The pattern was similar in a preliminary analysis of Post-Schar School state and congressional district polls.
But in a post-election analysis last week, Quantus Insights lead pollster Jason Corley said weighting to recalled vote was counterproductive:
The out-party’s low-propensity voters, the ones the voter file says probably won’t show, show up. And the in-party’s softer voters stay home or quietly drift in the last two weeks. Every tool we use to make a sample look like an electorate pushes against both of those. A likely-voter screen built on past turnout can’t see a surge electorate, because by definition those people didn’t vote last time. Anchoring on 2024 presidential recall pulls the sample toward an electorate that didn’t come back.
How do pollsters judge the performance of midterm election polls this year?
Rivers (YouGov): “Okay, not great. While we underestimated the size of the blue wave, nobody paying attention to the polls should have been surprised that there was a blue wave.”
Carvalho (Marist): “Excellent. Polls provided a solid narrative regarding the intensity and motivation of voters to support Democratic candidates along with the context that neither political party nor their nominees were particularly popular.”
Kimball (Emerson): “Good. They captured the broader trends and signaled where a Democratic wave could make races competitive.”
Worth your attention
Democrats renew faith in democracy, while Republicans lose it — In August, 73 percent of Republicans in an Economist-YouGov poll said democracy was working at least “somewhat well” in the United States compared with only 23 percent of Democrats. After last week’s midterm elections, Democratic faith in American democracy jumped to 44 percent while Republicans’ dropped to 51 percent. Read more.
Death penalty support at lowest point in 50 years, new Gallup poll finds — Support for capital punishment in the United States has reached its lowest level in more than 50 years, with 49 percent of adults saying in a poll released Tuesday that they are in favor of the death penalty for convicted murderers. Americans were surveyed in early October, in the weeks following Tennessee’s botched execution of convicted murderer Christa Pike. Read The Post’s story.
Republicans split over nominating a Trump-like candidate in 2028 — A Reuters-Ipsos poll this week found 52 percent of Republicans and Republican-leaning independents want the party’s next presidential nominee to “take a different approach to governing as Trump,” while 46 percent prefer “a similar approach” to Trump. Read more.
What do you want to know?
Send your questions, comments and ideas to [email protected].
The post Why pollsters think surveys undershot Democrats’ midterm strength appeared first on Washington Post.
