Two weeks after Election Day, nearly every 2026 Senate race has a winner, so we can grade the polls. The short version: they got most winners right and missed in one direction. In the 24 Senate races with a final polling average, a called result and a Democrat and a Republican as the top two, our average picked the winner 22 times. But the results came in more Democratic than the polls in 22 of those 24 races, by an average of almost 5 points on the margin.
That's a different kind of miss from the ones people remember. In 2016, 2020 and 2024, the polls underestimated Republicans. This year they underestimated Democrats, and fairly consistently.
The numbers
Here's how our final polling averages compared with the results in Senate races, measured on the Democrat-minus-Republican margin. The figures are as of November 16; a few counts are still moving.
| Races | Average miss | Results vs. polls | Wrong winner | |
|---|---|---|---|---|
| All Senate races with polls | 24 | 5.2 points | 4.8 points more Democratic | 2 |
| Competitive (final average within 10) | 14 | 4.2 points | 3.5 points more Democratic | 2 |
| Less competitive | 10 | 6.7 points | 6.7 points more Democratic | 0 |
Alaska is left out: its count is still going and its ranked-choice tabulation runs November 18. So is Montana, where the race was between Republican Kurt Alme and independent Seth Bodnar, with the Democrat a distant third; a Democrat-minus-Republican margin doesn't describe that race.
For context, RealClearPolling found that its averages in toss-up Senate races missed by 3.5 points this year, against a 3.8-point average for Senate toss-ups from 2008 to 2024. By that measure, 2026 was an ordinary year for accuracy. What stands out is the direction: in the competitive races, the results beat the Democrat's polling in all but two (Kentucky and Louisiana).
Where the misses were biggest
Two of the most-watched races had the largest misses among competitive contests:
- Georgia: our final average was D+9.1. The result was D+18.1, a 9-point miss.
- Texas: the final average was D+3.2, a race most polls called a toss-up. The Democrat won by 12.1, a miss of almost 9 points. Texas was also one of the most-polled races of the year, so this wasn't a case of too little data.
Two races went to the other candidate than the polls pointed to:
- South Carolina: the final average was R+0.7. The Democrat won by 4.5.
- Mississippi: the final average was R+3.5. The Democrat won by under a point, one of the closest Senate races of the year.
In states that weren't close, the misses were larger but mattered less. Illinois (D+13.4 in the polls, D+25.6 in the result) and New Mexico (D+17.5 against D+26.5) were each off by 9 points or more. These races drew few polls, and a single partisan or low-quality survey can move a thin average a long way.
The misses toward Republicans were few: only Kentucky and Louisiana came in more Republican than their polls, Kentucky by about 3 points.
It wasn't only the Senate
Governor's races showed the same pattern. In the 33 governor's races with a final average and a result, the outcome was more Democratic than the polls in 27, by about 4 points on average. Our averages pointed to the wrong winner in Georgia, Kansas, Texas and Vermont. Vermont was the one wrong call in the Republican direction: the polls had the Democrat narrowly ahead, and the Republican won by more than 6.
Other analyses found the same thing. Pushbuild looked at 215 individual Senate and governor polls from the final three weeks and found that 168 of them understated the Democrat, by 4.1 points on average.
Why the polls missed
No one knows yet, and it will take months of voter-file data to answer properly. But several explanations are worth testing against the numbers:
- A late break. In a midterm, undecided voters often break against the president's party. If a share of undecideds moved in the final days, polls that ended a week or two out would miss it. Our averages give more weight to recent polls, but a final-weekend move would still be too late for most of them.
- Turnout. Pollsters have to guess who will vote. A model built on the 2022 or 2024 electorate would understate a party whose voters turned out at unusually high rates. The size of the miss in places like Texas and Georgia fits that story, but turnout data by group isn't in yet.
- Corrections for past misses. After underestimating Republicans in 2016, 2020 and 2024, many pollsters changed how they weight their samples. If those changes pulled their numbers toward Republicans and this year's electorate didn't need the adjustment, the polls would lean the other way.
- Thin polling. In less competitive races, a handful of polls (sometimes from campaigns or partisan groups) made up the whole average. That explains bigger misses in those states, but not the consistent direction in the competitive races, which drew plenty of polling.
My own read: I'd put the most weight on turnout. When 22 of 24 races miss in the same direction, that isn't random noise. It points to an assumption almost every pollster shared, and the likeliest one is who they expected to show up. A late break would also fit, but I'd expect it to show up more in the closest races than in the safe ones, and the misses were, if anything, bigger in the safe ones.
What it means for reading polls
The lesson isn't that polls are broken. They picked 22 of 24 Senate winners, and their average error in competitive races was close to the historical norm. The lesson is that the error in one cycle tends to run in one direction across many races at once. A 3-point lead in the polls means less than it seems, because if the polls are off in your state, they're probably off by about as much, and in the same direction, in the states next door.
That's worth keeping in mind when the first 2028 polls start to come in.
How we measured this
We compared each race's final polling average (built from the general-election polls of the race in the 30 days before Election Day; primary polls are left out) with the result on the latest count, both as the Democrat's margin over the Republican. Races without polls, or without a Democrat and a Republican as the top two, are left out. The full table, including House and governor's races, is on our poll accuracy page, and each pollster's record is on pollster ratings. Results can still change by a few tenths as states certify; we'll update this piece if any change affects what it says.
Sources
- RealClearPolling: How the 2026 Polls Stack Up Against Recent Elections (Nov. 9, 2026)
- Pushbuild: How accurate were the 2026 midterm polls? (Nov. 7, 2026)
Written and edited by Brian Davis. Research help from AI tools; every number checked against the linked data. See a mistake? Tell us. Last updated .
Was this useful?
Comments
No comments yet. Be the first.
See an error on this page? Let us know · Missing something? Suggest it



