This week’s midterm elections were a big test for pre-election polls, which have made substantial changes to their methods over the past decade following significant election misses. So how did they perform?
Nationally, poll averages leading up to Election Day showed Democrats leading by roughly nine percentage points in support for Congress, making them strong favorites to flip control of the U.S. House even with Republicans’ redistricting advantage. But even that big lead was an underestimate, with the New York Times estimated that Democrats will win the popular vote by 13 percentage points, the largest margin for either party since 1976.
A similar pattern played out in seven toss-up U.S. Senate contests: Poll averages showed Democrats ahead by five points or less, but the party ended up winning six of these states by margins from four and 12 points, with Alaska still counting ballots. Polls came within two points of the current vote margins in Alaska, Kansas, Maine, Ohio and Michigan, a strong performance considering that pollsters have either struggled or rarely conducted polls in these states in recent elections.
But polls were far less accurate in Texas, underestimating Democratic Senate nominee James Talarico by nine percentage points, and in Iowa, where polls underestimated Democrat Josh Turek’s winning margin by four points. Democrats also outperformed the polls in states that handicappers rated as leaning toward one party — contests that generally attract fewer polls — with Democrats pulling off an upset in South Carolina and running up the scores in Minnesota, New Hampshire and North Carolina.
Want this in your inbox?
Sign up for The Pollwatcher newsletter here.
Preliminary results suggest that national Senate polls had one of largest systematic errors of recent election cycles, as the chart below shows. The 3.9-point average overestimate of Republican support is in between 2020 and 2024, when in which polls overestimated Democrats by 4.7 points and 2.9 points, respectively. And 2026 is the first election since 2012 that polls have overestimated Republicans by more than a percentage point on average.
Why did polls overestimate Republican support?
Pollsters will be digging into their own data for answers to this question over the coming months, but the election results point to a few possibilities. Polls showed Democrats were more interested in voting, but modeling likely voters is not precise, and some polls showed a sizable range of estimates across different turnout scenarios.
Relatedly, Democrats were significantly more likely than Republicans to answer polls, which can be a sign of higher turnout interest but also a warning sign for bias. In our own Kansas poll, Democrats were more than twice as likely to complete the survey after answering an interviewer’s call, and the survey found signs that President Donald Trump’s 2024 voters were underrepresented. The poll found Democrat Adam Hamilton leading by nine points, and he ended up winning by five.
Polls got the big picture right
Statistical accuracy is important, but it’s also important that polls paint an accurate picture of the mood of the country, reliably showing how support is shifting across different groups and the range of likely outcomes heading into Election Day.
All year, polls showed voters were frustrated about the economy and gas prices, which were exacerbated by an unpopular war with Iran. Trump’s approval ratings had fallen to midterm-cycle lows not seen since President Harry S. Truman, and polls showed Democrats with a rare advantage on trust to handle the economy. Election Day polling seemingly confirmed most of these patterns, with two-thirds of voters saying Trump’s policies have worsened economic conditions and more voters trusting Democrats than Republicansto handle the issue.
Polls also correctly forecast how different groups shifted, according to an analysis by TrueDot pollster (and former Post polling director) Jon Cohen. He found pre-election polls correctly forecast Democrats winning young voters by more than 30 points and winning Latino voters by a wide margin, both shifts from 2024 that were mirrored in post-election data.
Veteran Pew Research Center pollster Scott Keeter offered measured praise for polls’ performance while saying a more detailed analysis is on the way.
Scott Keeter, Pew Research Center senior survey adviser
Our early observations are that the pre-election polls did a good job telling the story of the electorate’s mood this year: deep dissatisfaction with the president – particularly his economic policies, including among some of his 2024 voters. Polls also consistently identified a wide “enthusiasm gap” – with Democratic supporters more interested, engaged – and likely to show up – than Republican supporters. It appears that pre-election polls did understate Democratic support in many of the major statewide races for governor and U.S. Senate, though poll watchers are still assessing the magnitude of the phenomenon. Stay tuned for more details as we finish our preliminary analysis.
Election polls also accurately laid out the likely scenarios ranging from a strong Democratic night where they flipped the House to a tsunami where they won in a slew of strongly Republican states. The actual result was hard to imagine of course, but polls pointed to the possibility.
Worth your attention
Ten key factors behind Americans’ midterm votes — A NORC survey of more than 90,000 midterm voters sponsored by The Washington Post and other organizations found Democrats made significant gains across a wide range of demographic groups since 2024, including Hispanic voters and young voters. And President Donald Trump’s high disapproval ratings — including in many states he won comfortably in 2024 — helped Democrats flip several key Senate seats on their way to taking control of the chamber. Read the story
More high-quality estimates for how different groups voted — The network exit poll used to be the only source of data on how different groups voted, but this year brings a true feast for poll watchers. In addition to the NORC survey noted above, the AP and five TV networks produced a massive voter poll that incorporates exit polls and interviews with early voters, while the New York Times combined its pre-election polls and election models to estimate how different groups voted. Beyond survey-based estimates, The Post’s crack data and graphics teams broke down how different groups swung in 2016 based on the demographics of counties and congressional districts.
What do you want to know?
Send your questions, comments and ideas to [email protected].
The post The Pollwatcher: How accurate were the midterm polls? appeared first on Washington Post.
