Every cycle the numbers arrive early and loud. This year they show Democrats ahead on the generic ballot by six to eleven points, depending on the survey and the day. Senate races in red states sit inside the margin of error. The picture looks like a wave. The question is whether the samples behind those numbers match the people who will actually vote on November 3.
How the Samples Are Built
Most public polls now rely on a mix of text-to-web panels, online opt-in respondents, and a smaller share of live phone calls. Response rates have collapsed. The people who still answer tend to be more politically engaged, more educated, and more willing to talk to a stranger about politics. In practice that has produced samples in which high-propensity Democrats show up far more often than high-propensity Republicans. One set of interviews across several districts found reliable Democratic voters (those who had voted Democratic in each of the last four elections) running 129 percent above the expected share, while the comparable Republican group ran only 2 percent above expectation.
Mike Johnson: “When I’m in these big arenas, I ask for hands — ‘Show me your hand if you’ve answered a poll in the last six months.’ There will be like three hands among 300 people. I’m telling you, these polls are off. All of them.” https://t.co/j3kkBowFSC
— Aaron Rupar (@atrupar) October 25, 2026
When pollsters try to correct for this, they weight the sample so the share of respondents who say they voted for Trump in 2024 matches the actual 2024 result. That fix assumes the Republicans who answer the survey are similar to the Republicans who do not. The evidence says that assumption is shaky. Some Republican pollsters report a clear mode split: live calls produce more Republican-leaning results than the text or online portions of the same survey, even after the samples are matched on demographics and party registration. One experiment in Iowa showed online and live-call results nearly identical, while the text-to-web portion of the same effort had Democrats ahead by twelve points.
🚨 NEW: Polls indicate that the Democrats will win a a majority of the battleground Senate races in next weeks midterms
Alaska: DEM GAIN
🔵 51%
🔴 49%Georgia: DEM HOLD
🔵 54%
🔴 46%Iowa: GOP HOLD
🔴 47%
🔵 45%Kansas: GOP HOLD
🔴 48%
🔵 47%Maine: DEM GAIN
🔵 50%
🔴 48%Michigan: DEM HOLD
🔵 47%
🔴 46%North Carolina: DEM GAIN
🔵 57%
🔴 43%Ohio: DEM GAIN
🔵 47%
🔴 44%Texas: DEM GAIN
🔵 51%
🔴 49%Control of the US Senate: DEM GAIN
🔵 DEM: 52 (+5) (including 2 independents)
🔴 GOP: 48 (-5)[Via NBC aggregate of polling from ActiVote, Data for Progress, InsiderAdvantage & Fox News]
— Politics US (@PolitlcsUS) October 26, 2026
What the Recent Numbers Actually Contain
A New York Times/Siena survey of likely voters found Democrats leading 52 to 41 on the House ballot in respondents’ own districts. CBS/YouGov and YouGov national surveys have shown Democratic advantages in the eight-to-nine-point range on the generic ballot. Other firms land closer to six. Those leads are real in the data that was collected. They are not necessarily real in the electorate that will cast ballots. The same surveys that produce large Democratic margins also show that the respondents who identify as 2024 Trump voters are more likely than average to say they now disapprove of him or plan to vote for a Democrat. If the Trump voters who refuse to answer surveys are less likely to have switched sides, then weighting up the ones who do answer inflates the size of any Democratic swing.
Pollsters who have spoken publicly about the problem note that Republicans remain harder to reach and less willing to participate when they believe the process is stacked against them. That pattern has been present since at least 2016. Some firms have adjusted their methods to compensate. Many still have not, or their adjustments rest on the untested claim that the reachable and unreachable Republicans think alike.
What the Record Shows So Far
In the three presidential elections of the Trump era, the final public polls overstated the Democratic margin by an average of nearly four points. Midterm and off-year elections have not shown the same consistent bias. The 2018 and 2022 midterms, and the 2021 and 2025 state elections, produced smaller or opposite errors. That history is why some analysts treat the current Democratic leads as more reliable than the presidential-year misses. It does not erase the sample problem. If the voters who skip surveys are disproportionately the ones who will still show up on Election Day and vote Republican, the published margins will shrink once the ballots are counted.
The polls are not invented. They are measurements of the people who chose to answer. Those people are not a random draw of the likely electorate. Until the samples match the turnout, the numbers describe a different contest from the one that will decide control of the House and Senate.