How each midterm pollster differs from the field

YouGov has published more polls of the 2026 midterms than any other firm: 105 general-election polls of Senate, governor and House races and the generic congressional ballot, according to our database. Public Policy Polling is next with 103, then the University of New Hampshire with 82, Emerson College with 77 and Siena College with 67. Michigan's Glengariff Group (59), Quantus Insights (58) and co/efficient (50) complete the list of firms with 50 or more.

Volume is not the same as accuracy, and it is not the same as agreement. We measured how far each pollster's figures this year sit from everyone else's, how each has done at past elections, and how much our forecast changes their numbers before they enter the averages.

How far each pollster sits from the field

A house effect is how a pollster's figures usually differ from other pollsters'. To measure it for 2026 we compared every general-election poll in our race averages with the polls of the same race by other firms ending within 21 days, and every generic-ballot poll with other firms' polls ending within 10 days. That gives 727 comparisons across 122 pollsters, on the Democratic-minus-Republican margin of the decided vote, using the figures as published. Idaho's Senate race, where the main challenger is an independent, is left out.

Most of the busiest firms sit close to the middle. YouGov's 78 comparable polls put the Democratic margin 0.8 points above other pollsters on average. Siena's 35 are level with the field, and Emerson's 28 are 0.1 points to the Republican side.

The clearest gaps are elsewhere. co/efficient's 25 polls have put the Democratic margin 4.7 points below other firms polling the same races in the same weeks, and RMG Research's 17 polls 3.9 points below. Echelon Insights sits 1.8 points to the Republican side across 20 polls, Trafalgar Group 1.5 points across 15 and Rasmussen Reports 1.2 points across 26. On the other side, Marist College's 16 polls have the Democratic margin 4.0 points above the field, Quinnipiac University's 12 polls 3.5 points above and Change Research's 11 polls 4.1 points above.

A house effect is not an error. Until the votes are counted nobody knows which side of the field is closer. In New Jersey and Virginia last November, every one of the 39 polls in our database that ended in the final month understated the Democratic margin: by 8.9 points on average in New Jersey and 5.7 in Virginia. Firms with a Democratic house effect looked better there. That says nothing about whether they will again.

Who has the record

Our ratings score every pollster on its final polls of Senate, governor and presidential races since 2000, and on how openly it publishes. Of the 502 pollsters we rate in the US, Marquette University is first on 85 out of 100, a grade A, with its final polls missing each party by 1.2 points on average at eight elections. Echelon Insights is fourth on 79, also an A, missing by 1.0 points at 10 elections.

The busiest firms mostly sit in the middle. Emerson scores 68 (B) from 111 elections and an average miss of 2.1 points per party. YouGov scores 63 (B) from 128 elections and a miss of 3.0 points; at those elections its final polls had Democrats 2.0 points too high on average and Republicans 1.4 too low. Siena College, often treated as a gold standard, scores 54 (C): a 2.7-point miss over 77 elections, and 16 out of 30 for openness.

The firms with the largest house effects this year tend to have thinner or weaker records. co/efficient scores 44 and RMG Research 40, both grade D. Quantus Insights, with 58 polls this cycle but only two past elections to judge, also scores 40. Marist (62) and Quinnipiac (61) both rate B.

Trafalgar Group shows the difference between a lean and a miss. At 50 past elections its final polls had Republicans 1.7 points too high and Democrats 2.3 points too low on average, and it scores 54 (C). The Silver Bulletin reported in January that Trafalgar's 2025 polls overstated Republicans by 12.6 points on average, the largest of any firm it tracked.

What our adjustments do to the averages

Our race averages do three things to a poll before it counts. They take off the pollster's house effect at past final polls, shrunk toward zero for firms with few races. They move polls paid for by a party, or from a firm that works mainly for one side, back by 2.5 points, the lean such polls have shown against independent polls of the same races across 556 final polls, and give each of them half weight. And in Senate and governor races they take 0.7 points off the Democratic margin, half the average past statewide miss. Polls are then weighted by recency, sample size and the pollster's record.

In most races the net effect is small. We compared each average with the same polls, weighted the same way but taken as published (after the move every poll gets for the national trend). The adjustments shift the Texas Senate average by less than 0.1 points, Maine's by 0.1 toward the Republicans and Iowa's by 0.3 toward the Democrats. Across the 45 Senate and governor races and House districts with six or more polls in their averages, the largest moves are between three and four points, in races where one party's pollsters do most of the polling.

The biggest move toward the Democrats is in South Carolina's Senate race: R+3.7 as published, R+0.7 adjusted, a move of 3.0 points toward Annie Andrews. Rasmussen, InsiderAdvantage and Trafalgar, all classed as working mainly for Republicans, carry 87 percent of the weight there. The biggest moves the other way come where Democratic-sponsored polls dominate: 3.8 points toward the Republicans in Kentucky's 6th district and 2.9 in the Oregon governor's race. In Alaska's Senate race the polls average R+2.2 as published and R+0.3 adjusted, a move of 1.9 points toward Mary Peltola. Polls from Rasmussen, Trafalgar and co/efficient, all classed as working mainly for Republicans, carry 49 percent of the weight there. In Georgia the move is 1.7 points, from D+8.6 to D+10.3. Trafalgar's 9 October Georgia poll, D+8.5 as published, counts as D+13.7.

That is where this year's evidence and the past record part company. For Trafalgar, its past house effect and its partisan classification together move each poll 6.1 points toward the Democrats. Against other firms this year, its polls sit 1.5 points to the Republican side. Our model leans on the past record because it is measured against results, not against other polls. The cost is that a firm which has changed its methods can be corrected for an old lean.

The wider margin of error

These adjustments move key Senate averages by up to three points. Past misses have been larger than that, in both directions. Final polls of 806 Senate, governor and presidential races since 2000 missed the result by 4.6 points on the margin on average, according to our track record. In 2020 they overstated the Democrats by 5.8 points across 82 races; in 2012 they understated them by 2.2; last year in New Jersey they understated them by 8.9.

Our generic-ballot average, adjusted the same way, stands at D+8.8 from 17 pollsters. Twenty days before the vote, the question for every pollster in this piece is the same one it faced in 2025: whether the field as a whole, not any single firm, is in the right place.