Every election ends with a scramble to name the "most accurate pollster." Most of those lists aren't worth much, because a pollster that got two races right can top the table by luck. So this piece looks first at the firms that polled a lot of races, then at what the whole field had in common: in 2026, almost all of them missed in the same direction.
How we graded them
For each pollster, we took its final poll in each race (taken within 30 days of the election) and compared it with the result, as the Democrat's margin over the Republican. We counted every Senate, governor and House race called between a Democrat and a Republican. A pollster needs at least two races to be ranked. The full table of 78 ranked firms is on our pollster ratings page. The numbers below are as of November 16. Montana's Senate race, where an independent finished second, and primary-season polls are left out.
The firms that polled the most
Thirteen pollsters had final polls in 10 or more races. That's enough to say something about them, though still not a lot:
| Pollster | Races | Right winner | Average miss | Results vs. their polls |
|---|---|---|---|---|
| CNN/SSRS | 12 | 9 of 12 | 3.2 | 2.8 more Democratic |
| AtlasIntel | 18 | 18 of 18 | 3.5 | 2.1 more Democratic |
| Public Policy Polling | 12 | 8 of 11 | 3.5 | 1.6 more Democratic |
| New York Times/Siena | 25 | 23 of 25 | 4.1 | 2.9 more Democratic |
| Echelon Insights | 11 | 9 of 10 | 4.6 | 4.3 more Democratic |
| YouGov | 21 | 19 of 21 | 5.1 | 4.7 more Democratic |
| University of New Hampshire | 12 | 10 of 11 | 5.4 | 1.3 more Democratic |
| Emerson College | 28 | 22 of 28 | 5.6 | 5.1 more Democratic |
| Research Co. | 21 | 20 of 21 | 5.9 | 5.5 more Democratic |
| InsiderAdvantage | 13 | 9 of 12 | 6.2 | 6.2 more Democratic |
| Marist | 11 | 9 of 10 | 6.2 | 6.2 more Democratic |
| ActiVote | 14 | 12 of 14 | 6.6 | 6.5 more Democratic |
| Trafalgar Group | 10 | 6 of 10 | 8.0 | 8.0 more Democratic |
("Right winner" leaves out races where the poll showed a tie.)
A few things stand out:
- AtlasIntel picked all 18 winners and missed by 3.5 points on average, the best record among the high-volume firms. It also came close on the national generic ballot, according to the Zoose Political Index's pollster report card.
- The New York Times/Siena polled more races than anyone but Emerson and picked 23 of 25 winners. In Texas, its final poll had the Democrat up 7; he won by about 12, according to the Zoose Political Index.
- Emerson College polled the most races (28) and picked 22 winners. Its polls ran about 5 points too Republican on average, close to the pattern for the field as a whole.
- Trafalgar Group had the weakest record of the group, picking 6 of 10 winners with an 8-point average miss, all in the Republican direction. Its final Texas Senate poll had the race tied; the Democrat won by about 12, according to Pushbuild.
Smaller samples, bigger caveats
Among firms with five to nine races, Survey 160 stands out: it picked all 8 winners and missed by 2.0 points on average. The Zoose Political Index, using a different method, found a similar result for Survey160: an average miss of about 1.4 points across four Senate races. Suffolk University (7 for 7, 2.9 points) also did well.
The very top of our full table is made up of firms with only two races each, such as RedBox Research (0.6 points). That's too few to separate skill from luck, which is why we show the race count next to every rating.
Nearly everyone missed the same way
Of the 25 pollsters with five or more races, 24 had results come in more Democratic than their polls. The one exception, Concord Public Opinion Partners, ran 3.5 points too Democratic.
That matters more than the rankings. When almost every firm misses in the same direction, the problem usually isn't one pollster's method. It's something they share: an assumption about who would vote, how undecided voters would break, or how to weight their samples after years of underestimating Republicans. We look at those explanations in Where the 2026 Senate polls missed, and why.
It also means the firms near the top of the table aren't necessarily better at polling. Some of them may simply lean the way this year's electorate did. A pollster with a small, consistent Democratic tilt would look very accurate in 2026 and very inaccurate in 2024. Their record over several elections is a better test than one year.
What I'll look for before trusting a pollster in 2028 is a long record and an open method: how many races it has been graded on, and whether it publishes its sample size, field dates and how it weights its results. One lucky call in a big race means less to me than a steady record across 10 or 20.
What this means for reading polls
- Look at the race count. A pollster that's perfect in two races has told you very little.
- Look at direction, not just size. A firm that misses by 4 points in both directions is different from one that misses by 4 points always toward the same party.
- Trust averages over any one poll. Combining many polls smooths out each firm's quirks. It can't cancel out a miss that every firm shares, which is what happened this year.
How we measured this
Each pollster's final poll in each race (within 30 days of the election) is compared with the latest count as the Democrat's margin over the Republican. Races without a Democrat and a Republican as the top two are left out. Results can move by a few tenths until states certify; we'll update this piece if a change affects what it says. See pollster ratings for the full table, and poll accuracy for every race.
Sources
- Zoose Political Index: The 2026 Pollster Report Card (Nov. 16, 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



