Every Poll on the 2026 Midterm

How to read the numbers. "D+5" means the Democrat led or won by 5 points. "R+5" means the Republican did. A poll's "miss" is its final margin minus the result. A minus sign means the poll had Democrats too low. "LV" means likely voters, "RV" registered voters, "A" all adults.

Type labels used on every item: EXIT POLL (the 2026 network Voter Poll of people who voted, run by SSRS) · POST-ELECTION SURVEY (fielded on or after Election Day, or released after it with fieldwork running into Election Day, so some people answered before they voted) · FINAL PRE-ELECTION POLL (a pollster's last poll, judged against the result) · PRE-ELECTION SURVEY (issues/demographics) (earlier polls cited for why voters moved) · FOCUS GROUP · ESTIMATE/MODEL (not a poll).

1. The short version

The polls named the right winner in almost every big race, but they lowballed Democrats. The national House vote looks like about D+13, against final polling averages near D+9 (8.8 at RealClearPolitics, 8.9 at Silver Bulletin, 9 at FiftyPlusOne). That is a miss of about 4 to 5 points toward Democrats, the reverse of 2016, 2020 and 2024. The most accurate national pollsters were the ones willing to publish very large Democratic leads. Both public rankings put AtlasIntel (D+14.1) and Tufts/YouGov (D+12.0) at or near the top. G. Elliott Morris's ranking ties his own Strength In Numbers/Verasight poll (D+12) with them. NYT/Siena and RMG Research (both D+11) come next. Marquette (D+13, from its own post; poll dates not given) and Big Data Poll (D+12.3) were also close but are in neither ranking. Republican-leaning firms Rasmussen Reports and Echelon sat near the bottom. Morris wrote that "Republican firms fill the bottom," but not all did: Cygnal (Republican) was mid-table, and Big Data Poll, often grouped with Republican firms, came close. Trafalgar's state finals had Paxton, Collins and Graham ahead, and all three lost. On what voters cared about, most polls point the same way: prices and the cost of living at or near the top, government corruption right beside it, and a vote against Trump himself (37% to 38% approval). The Iran war was very unpopular and very visible, but no poll in our research shows it as voters' top issue. Polls suggest much of its effect ran through gas prices.

2. Accuracy scoreboard

2a. The yardstick: what "the result" means right now

No final national House count exists yet. This rundown grades every poll two ways.

• Against D+13, the working estimate. This is the figure G. Elliott Morris used for his ranking ("a predicted margin of 13 points for the Democrats as of 11:30 AM Nov 5"). It also matches Axios's report of The Argument's estimate ("roughly 13 points"), VoteHub ("about D+13") and our own analysis. It is an estimate, not a count.
• Against D+11.4, the only real count in our research with a share counted. AtlasIntel's chart quotes it: "The CNN count of all 435 districts stood at D+11.4, with about 89% of the expected vote, at Nov. 4, 10:11 a.m. ET." This count is incomplete. Slow states such as California were still counting, so it is expected to grow toward Democrats.
• Other estimates, all preliminary: AtlasIntel's projection D+13.8; NYT Times Voter Trends D+14; Morris D+14.3 (Nov. 4, 9:30 a.m.), with a range of 12 to 15; Will Jordan D+12.8; Blue Rose Research "~57%" of the two-party vote; Sean Trende about D+12.

Sources: HWStP1QWEAAuX7w.jpg (AtlasIntel on X); HWYB0AHWIAEdgyB.jpg (Morris on X); Axios; VoteHub on X; HWUUG43XwAETFY7.png; gelliottmorris-2118112747267203548-1.jpg; Jordan on X.

2b. Every pollster with a final national House number

Rows follow AtlasIntel's published order. Pollsters only Morris ranked are slotted in by their margin. The two "miss" columns are our arithmetic. The last two columns are the two published rankings exactly as printed. Field end dates and populations are from AtlasIntel's chart.

Final national numbers in neither ranking

Polling averages

Things in this table that don't line up - Morris's misses for three firms imply a margin about 1 point lower than AtlasIntel lists: Zogby (−8.0 implies D+5, AtlasIntel says D+6), Targoz (−11.0 implies D+2, AtlasIntel says D+3) and The Argument/Verasight (−3.0 implies D+10, AtlasIntel says D+10.8). Our guess is that the two charts used different versions or rounding of the same polls. Neither chart says. - Cygnal published three different versions of its final. Its deck text says "Democrats lead the generic ballot 51% to 43% (D+8)" (read from the saved file). Its Oct. 27 press release says "52 percent to 43 percent" (D+9). Both rankings use D+8.4. - Echelon's saved October omnibus (Oct. 15–19) shows 50 to 45 (D+5). AtlasIntel lists Echelon's last poll as ending Oct. 23, and Morris counts two Echelon polls. The Oct. 23 poll was not found. - AtlasIntel marks Tufts as a poll of all adults ("A"). Morris's chart says it only uses likely-voter surveys. - AtlasIntel left out Strength In Numbers/Verasight. AtlasIntel says "the final poll released by Verasight is correctly included," meaning The Argument/Verasight poll. Morris calls the exclusion "Very dishonest." Morris, for his part, runs Wave Polling and Research, which also appears in his ranking.

2c. Key races: final polls against the count

Results are from NBC's Senate results page (Nov. 5, 2:03 p.m. ET, via the combined report) unless marked. Governor results are AP counts via Newsweek (Nov. 4, 2 p.m. ET) unless marked. Misses in parentheses are poll minus result. Where a miss is our own arithmetic rather than the source's, the notes treat it as preliminary.

Sources: Newsweek (RCP vs. AP); The Hill (DDHQ); HWKVjCxX0AAGKmY.jpg and HWS1csMXEAA3CVK.jpg (AtlasIntel); Nate_Cohn-2118051865908912461-1.jpg (NYT/Siena, Cohn on X); the trafalgar folder; PatrickRuffini-2117821937162314000-1.jpg (ElectIndex averages, Ruffini on X); Marquette on X; NBC Senate results.

What the race table shows - The averages missed toward Republicans in every race listed except Alaska, which is still being counted, by 2 points (Ohio Senate) to 9.5 (Texas Senate). Two averages had the wrong winner: Texas governor and South Carolina Senate. - AtlasIntel and NYT/Siena were not biased in one direction in Senate races, unlike the averages. Across the eight Senate races where we have both its poll and a count, AtlasIntel's misses average about +0.2 in direction and about 2.9 points in size. Across the seven where NYT/Siena polled, its misses average about −0.4 in direction and about 2.1 points in size. On the seven races both polled, NYT/Siena's average miss was smaller than AtlasIntel's. This is our arithmetic on preliminary counts, and Alaska is only 75% counted. - Trafalgar missed every race toward Republicans, by about 7.1 points on average across its five Senate finals (our arithmetic). It had the wrong winner in Texas, Maine and South Carolina.

2d. The published rankings and error measures

Saved files: HWStP1QWEAAuX7w.jpg; HWS1csMXEAA3CVK.jpg; HWYB0AHWIAEdgyB.jpg; HWYbEeYXQAA5v1l.jpg; HWYbEeoXkAEEIUI.jpg. Source posts: AtlasIntel ranking; AtlasIntel scorecard; Morris ranking; Morris error table.

3. The Iran war: what every poll says

The question. Did any poll show the Iran war was voters' number one issue? Or was it one part of a bigger story?

3a. Every measure of the war in our research

Claims about the war with no published poll behind them

3b. The war in the state exit polls

All from the CNN state files (stamped Nov. 4, 12:07 to 3:56 p.m. ET; Arizona is older, 2:36 a.m. ET, "stage 7"). Read from the saved files. "Crucial" percentages are the share of voters who called each issue crucial. The vote shown is how the war-crucial voters split, D-R.

Saved files: cnn/api folder (for example 2026-SG-TX.json, from CNN). Kansas and South Carolina had about 1,000 respondents and no in-person interviews, so their numbers carry wider error.

In every one of these races, the voters who called the war crucial backed the Democrat at least as strongly as the voters who called the cost of living crucial (our check of the same files).

3c. Verdict

No poll in our research shows the Iran war as voters' number one issue. Where a poll let voters pick one top issue, foreign policy or the war got 1% to 6%, far behind the economy and the cost of living. Where voters rated issues as "crucial," the war came third or fourth (34% nationally), behind government corruption (55%) and the cost of living (49%). That held in all 14 state exit polls checked. But the war was not a minor factor. Most voters (63%) called it the wrong decision, the same share that disapproved of Trump. It was the second-most-heard-about issue before the vote (57% "a lot," Data for Progress). Before the election, voters expected it to be the second-biggest reason if Republicans lost (51%, behind the cost of living at 65%, Echelon). Two caveats matter. First, the third of voters who called the war crucial broke harder for Democrats (72-26) than any other "crucial" group. That shows who cared about the war, not that the war switched their votes; many were likely Democrats already. Second, much of the war's effect seems to have run through gas prices, though no poll measured that directly. Voters picked ending the war over restarting it mainly to bring gas prices down (Fox/NORC), named "ending the war in Iran" as a top fix for the cost of living (74%, Echelon), and linked "the endless wars" to "high inflation" in their own words (Echelon). The data fit a simpler read: the war was a real part of the price story, not the top issue on its own. Vance's claim that it was "the reason" rests on a pollster he has not named.

4. Pollster by pollster

Ordered from most to least accurate on the national House vote, using the two published rankings. Pollsters with no national number come after. Pollsters with only a number in our research are grouped in short tables.

4.1 AtlasIntel

Who they are. A polling firm led by CEO Andrei Roman. It uses "Atlas Random Digital Recruitment," which recruits people "randomly during routine web browsing" on any device and then weights them by "gender, age group, education level, income level, region, and previous electoral behavior" (read from the saved national PDF). Its state and Latino polls carry a PollsterGraph co-brand. The notes give no partisan lean or sponsor. Axios: "It's the second straight election where the pollster stood out for its accuracy." Roman claims "being the top US pollster for 4 consecutive cycles is unprecedented in US polling history," and that AtlasIntel was most accurate in 2026 in Hungary and Colombia too.

Accuracy grade. National: top tier in both rankings. Its D+14.1 sat 1.1 points above the D+13 estimate, or 2.7 above the incomplete D+11.4 count. States: mixed, but not biased in one direction. It was exact in Iowa Senate and Ohio governor and within 0.1 in Michigan. It was well short on Texas (−6.2) and North Carolina (−4.7), and too high for Democrats in Alaska (+5.2, partial count), Ohio Senate (+2.7) and Kansas (+2.0).

Item 1. FINAL PRE-ELECTION POLL: national House ballot. - Field Oct. 28 to Nov. 1; 3,566 likely voters; ±2. Posted Nov. 1, 9:00 p.m. ET. - Top line: "Democratic candidate: 55.5% / Republican candidate: 41.4% / Don't know: 2.3% / Would not vote: 0.8%." (D+14.1) - Topics in the PDF: presidential approval, generic ballot, key issues, image of political figures, economic perceptions, trust in institutions, 2028 primaries. The PDF's numbers are in chart images and were not extracted (gap). - Its own claim, Nov. 3, 10:55 p.m. ET: "Since the beginning of 2026, AtlasIntel was the only pollster to project a generic ballot lead for democrats of more than 14pp." - Saved: 2026_midterms__national_poll_110126_1_87e74540c4d7f251.pdf; HWFbbUaXYAAZb1e.jpg. Source: AtlasIntel on X; PDF; AtlasIntel polls page.

Item 2. FINAL PRE-ELECTION POLLS: 10 Senate races. Field Oct. 28 to Nov. 1, likely voters, "Share of all respondents." Posted Nov. 2, 8:01 p.m. ET: "Democratic candidates lead in all 10 U.S. Senate races surveyed by AtlasIntel."

Saved: HWKVjCxX0AAGKmY.jpg, the state PDFs in the atlasintel folder, and 2026_midterms__texas_110126_1_e8958087c4dc49a8.pdf (Texas, Oct. 28–Nov. 1, 1,678 likely voters, ±2). Source: AtlasIntel on X.

Item 3. FINAL PRE-ELECTION POLLS: governors (from its scorecard). Ohio, Acton D+7.9 vs. result D+7.9 (0.0). Iowa, D+10.1 vs. D+11.0 (0.9). New Hampshire, R+5.9 vs. R+3.8 (2.1). It also says it "projected a Democratic gubernatorial victory" in Texas; the number is not in our research. Claim: "6 / 6 Governor winners correctly called."

Item 4. POST-ELECTION SURVEY: Latino poll (released Nov. 4). - 878 likely Latino voters; field Oct. 26 to Nov. 3; ±3. Caveat: fieldwork ran through Election Day and the vote question is worded in the future tense ("Which party will you vote for"), so this is mostly a pre-election poll released after the vote. - House vote: "The Democratic candidate 61.6%", "The Republican candidate 37.8%", "Don't know 0.6%" (D+23.8, the lowest of the five Latino estimates; the others run D+27 to D+40). - Trump approval among Latinos: approve 31.1%, disapprove 67.9%, don't know 0.9%. - Groups (Democratic share): men 61.2, women 69.8; ages 18–29 93.0; 30–44 43.4 (56.6% Republican); 45–64 56.6; 65+ 84.0; college 63.3, no college 59.5; Mexican-origin 72.9; Puerto Rican 50.9; Cuban-origin 14.6 (85.4% Republican); Dominican 88.9; Salvadoran 75.0; Venezuelan 52.3; Colombian 40.9; 2024 Trump voters 16.1 (83.9% Republican); 2024 Harris voters 98.6; 2024 non-voters 87.2. Flag: the age pattern (93% D among the young, 57% R among 30–44, 84% D among 65+) is unusual. These are small subgroups. Hold them. - Issues ("most important in determining your vote," up to 3): immigration policy, deportations and ICE 40.7%; combating corruption 39.6%; abortion and reproductive rights 37.8%; economic policy and taxes 29.5%; inflation/cost of living 23.2%; positive influence in Latin America 23.1%; safeguarding democracy 22.2%; crime 18.4%; racial discrimination 14.0%; healthcare 12.3%; general wellbeing of Latinos 11.1%; religion and moral values 8.6%; democratic transition in Cuba, Venezuela and Nicaragua 5.4%. - Trump policies, success vs. error: fossil fuels 48/28 (+20); "only two genders" 52/42 (+10); mass deportation 45/40 (+5); federal layoffs 43/46 (−3); China trade war 37/44 (−7); tariffs 39/48 (−9); bombing Iranian nuclear facilities 44/55 (−11); pressure on Powell 33/45 (−12); DEI pressure 42/55 (−13); tax-cut bill 39/57 (−18); environmental deregulation 38/58 (−20); Medicaid cuts 30/61 (−31). - What AtlasIntel said, quoted exactly: "In today's election, most of these 2024 gains were reversed due to poor presidential approval, which is considerably lower among Latinos when compared with the national average." "During the 2026 midterms, the top of mind issue for Latinos was immigration, specifically deportations and ICE raids." "64% of Latinos believe Trump's administration policies are biased and unfair against them." "Most Latinos approve of Trump's Venezuela intervention, but it was not a key driver of their voting decisions." "More than 60% of Latinos who voted in the 2026 midterms chose the Democratic House candidate ... Young Mexican-origin Latinos strongly drove this." "About a third of tonight's surprise comes from the Latino vote." - Conflict worth knowing: AtlasIntel's Latino issue ranking (abortion 37.8%, inflation 23.2%) is very different from UnidosUS's (cost of living 58%, abortion 3%). The questions and answer lists differ, so they are not directly comparable. - Saved: us_midterms__latino_poll__eng_110426_088f4b0160bbed13.pdf (charts are images; numbers read from the saved slides HWQX668W0AA0kZ6.jpg, HWQZIFKXEAARwya.jpg, HWQZ_RuWgAEKuxg.jpg, HWQcABSXQAAsH-3.jpg). Source: PDF; AtlasIntel on X; X; X; X; X.

Item 5. Self-published accuracy scorecards (not polls). National ranking: "AtlasIntel was the most accurate pollster of the 2026 Midterms. Please note: the ranking is unlikely to change but exact errors depend on the final vote count." State scorecard: "#1 / 32 Generic ballot accuracy," "10 / 10 Senate winners correctly called," "6 / 6 Governor winners correctly called." Problems: it grades against its own projection; it shows only its seven best state races; Alaska was still uncalled on Nov. 5; and its line that in Texas it "was one of only two pollsters to foresee a Democratic Senate win" is hard to square with the RCP average (Talarico +2.6) and NYT/Siena (Talarico +7); the claim may mean something narrower. Reply to Morris: "The comparison includes the last and final poll published by each polling firm ... the final poll released by Verasight is correctly included." And: "the current generic ballot estimate updated by the NYT is D+14, not D+13." Sources: ranking; scorecard; reply; reply.

Item 6. POST-ELECTION SURVEY (type unconfirmed): "Atlas VRC" ad test with Latinos. Released Nov. 5, 1:12 p.m. ET; field dates and sample not given, so it may have been fielded before the vote. Finding: "Maria Elvira Salazar's ad criticizing immigration policy and ICE raids was highly successful, while Trump's response to it was badly received." Roman: "Salazar's 30-second ad denouncing immigration policy saved her seat. Trump's response to it accelerated the Latino swing nationwide." Sources: AtlasIntel on X; Roman on X.

The dispute. Morris: "AtlasIntel has deliberately excluded Strength In Numbers/@verasight_data from their analysis of pollster accuracy. Depending on whether you use a benchmark natl vote of D+13 or D+14, we are either all tied for first or coming in at 1 and 2. Very dishonest." Roman: "I find it troubling that after being the most precise pollster 4 election cycles in a row, AtlasIntel still gets quite low coverage in US media." Axios: "AtlasIntel placed itself first in a ranking of final national House polls based on its projection of the popular-vote margin." Sources: Morris on X; Roman on X; Axios.

4.2 Tufts University (Tufts/YouGov)

• FINAL PRE-ELECTION POLL. D+12.0; field ended Oct. 27; AtlasIntel marks it as all adults. AtlasIntel #2 (error 1.8). Morris tied #1 (−1.0). Miss vs. D+13: −1.0.
• Nothing else about this poll is in our research. (Tufts's CIRCLE youth-vote survey is a separate project; see section 6.)

4.3 Strength In Numbers / Verasight (G. Elliott Morris)

Who. G. Elliott Morris writes Strength In Numbers and runs the FiftyPlusOne average. Our research describe him as CEO of Wave Polling and Research, whose MRP polls rank #10 on his own chart. No partisan lean is given in our research.

• FINAL PRE-ELECTION POLL (October; exact dates and sample not in our research). Morris: the final reading was "D+12 among likely voters two weeks ago, and D+14 if you looked only at the behavior of people who said they were definitely going to vote." Also: "58% of adults said it's more important to elect a Congress that acts as a check on President Donald Trump, versus just 25% who want one that helps him pass his agenda." Morris's ranking: tied #1 (−1.0). Left out of AtlasIntel's ranking. Source: Strength In Numbers, Nov. 4.
• Post-election analysis (not a poll). Headline: "Democrats win U.S. House, Senate in biggest incumbent-party defeat since Watergate." In MS NOW, using the MS NOW Voter Poll (the same SSRS exit poll): "the Republicans lost the most ground with Latinos, young people and lower-income Americans"; Latinos gave Democrats "a whopping 27-point margin," versus a Harris margin of 3 in 2024 per Pew; rural voters "have moved 22 points left since 2024." On the polls: "What has surprised most election observers is not the size of the error, however, but the direction. Before last night, polls had overstated support for Democrats in four of the last five elections." And: "Republican firms fill the bottom, having underestimated Democrats significantly." Sources: MS NOW; Morris on X.
• Pre-election warning (ESTIMATE/MODEL). Oct. 16: "If Democrats beat today's polls by a normal margin, they'd win 56 Senate seats — including South Carolina and Nebraska." Source: Morris on X.
• Saved: strength-in-numbers-verasight folder (7 images).

4.4 New York Times / Siena College (Nate Cohn)

• FINAL PRE-ELECTION POLL: national. D+11.0; field ended Oct. 19; likely voters. AtlasIntel tied #3 (2.8). Morris tied #4 (−2.0).
• FINAL PRE-ELECTION POLLS: Senate. Cohn's table (Nov. 4, 1:44 p.m. ET): final polls Michigan D+11, Alaska D+1, Iowa D+1, Kansas D+5, Maine D+6, Ohio D+6, Texas D+7. Average 5.3, against his estimated results averaging 5.4. "Old weights" would have averaged 8.7. Cohn: "our final polls did not skew toward either side. Our old weighting approach would have overestimated Dems by 3. The lesson: nonresponse was probably a bigger problem than in prior midterms, but weighting to a 2024 electorate (like many other pollsters) obviously wasn't a solution, either." Saved: Nate_Cohn-2118051865908912461-1.jpg. Source: Cohn on X.
• ESTIMATE/MODEL: Times Voter Trends (Nov. 4, preliminary). Not a poll; the Times' estimates of how groups voted. From a screenshot posted by AtlasIntel's CEO (the NYT page is paywalled and was not opened): all voters "D+14" (56% to 43%), a D+15 shift from 2024; Hispanic voters D+40 (69% to 29%), a D+29 shift; white voters R+4; men D+3; women D+23. Will Jordan's chart from the same data: ages 18–29 D+26 (2020), D+9 (2024), D+41 (2026); Hispanic D+30, D+11, D+39; Black D+82, D+73, D+83; white non-college R+28, R+33, R+19. Cohn: "right now, we're D69-R29 among Hispanic voters v. 64-34 in the exits." Saved: HWUyPwDWwAA8d_L.png; williamjordann-2117958032969859343-1.jpg. Sources: Roman on X; Cohn on X; Jordan on X.
• Accuracy grade. Second tier nationally (−2.0). Unbiased on average in Senate races, with misses both ways (Michigan +4.0 too Democratic, Texas −5.1 too Republican).

4.5 RMG Research (Scott Rasmussen)

• FINAL PRE-ELECTION POLL. D+11.0; field ended Oct. 13; likely voters. AtlasIntel tied #3 (2.8). Morris tied #4 (−2.0).
• Scott Rasmussen, Nov. 5: "it looks like our final Generic Ballot poll showing D+11 will end up being pretty good." Election night: "Turning into a pretty good night for pollsters and a great night for prediction markets." Nov. 4: "In our 10-10-80 nation, the 10 percenters on the left routed the 10 percenters on the right last night. The 80 percenters remain unrepresented and frustrated." Sources: X; X; X.

4.6 The Argument / Verasight (Lakshya Jain)

• FINAL PRE-ELECTION POLL. AtlasIntel lists D+10.8 (Oct. 23, likely voters; #6, error 3.0). Morris lists −3.0 (tied #7). The Argument says "Their final model had D+11 among likely voters."
• Jain: "polling error is not one-sided, and that polls can undershoot Democrats. Feels like we saw a lot of herding from pollsters who couldn't believe their data." Headline: "Republicans did even worse than we warned" (paywalled). Jerusalem Demsas, using The Argument's pre-election polling (PRE-ELECTION SURVEY), says voters saw Democrats as "compassionate" and "moderate" and the GOP as "extreme." Saved: lxeagle17-2117855665104535647-1.jpg. Sources: Jain on X; The Argument; The Argument, 9 takes.

4.7 Second tier: final national number only (D+10 to D+10.7)

All are FINAL PRE-ELECTION POLLS. Numbers from AtlasIntel's and Morris's charts.

4.8 YouGov

• FINAL PRE-ELECTION POLL. D+9.0; field ended Oct. 30; likely voters. AtlasIntel tied #12 (4.8). Morris #11 (−4.2, 4 polls). YouGov also fields the Tufts and McCourtney Institute polls, which are ranked separately.
• POST-ELECTION SURVEY, Nov. 4, 1,995 U.S. adults, ±2.4: results reflect voters' opinions of Trump 38%, the candidates on the ballot 11%, both equally 32%.
• POST-ELECTION SURVEY, Nov. 4, 1,905 adults, ±2.9: 80% said Democrats did better or won; 7% said Republicans. Among Republicans: 73% Democrats, 17% Republicans.
• POST-ELECTION SURVEY, Nov. 4, 2,421 U.S. adults (read from the saved image): "How united or divided is the country at present?" Very united 6%, somewhat united 13%, somewhat divided 24%, very divided 49%, not sure 8%.
• Saved: HWSXr09XQAAzV_0.jpg, HWTFSG7bsAAs6XQ.jpg, HWT-xMMaoAAXp6I.jpg. Sources: YouGov on X; X; X.

4.9 AP-NORC and the Fox News Poll (Beacon Research / Shaw & Co.)

• AP-NORC, FINAL PRE-ELECTION POLL. D+9.0; field ended Oct. 12; registered voters. AtlasIntel tied #12.
• Fox News Poll (Beacon Research, Democratic partner Chris Anderson; Shaw & Co., Republican partner Daron Shaw), FINAL PRE-ELECTION POLL. D+9.0; field ended Oct. 12; registered voters. AtlasIntel tied #12.
• Anderson: "Polls probably, on average, undercounted Democrats significantly." "This is sort of a seismic election and you can't model for a seismic election. We definitely didn't undercount Republicans, that's for sure."
• Shaw: "If we're talking about the direction of the polls, they were clearly correct." "we probably underestimated Democratic vote share by a couple of points." "I guess on Texas, the magnitude was surprising."
• Source: Fox News, Nov. 4.
• The separate Fox News/NORC AmeriSpeak survey on the Iran war (Nov. 1–3) is in section 3 and section 6.

4.10 Cygnal (Republican)

• FINAL PRE-ELECTION POLL. Oct. 23–24; 1,500 likely voters; ±2.53. Deck: "Democrats lead the generic ballot 51% to 43% (D+8)." Press release (Oct. 27): "Democrats lead the generic ballot 52 percent to 43 percent." Both rankings use D+8.4 (AtlasIntel #15, error 5.4; Morris #12, −4.6).
• What it found, from the deck (read from the saved file):
• Already voted: "backed Democrats by a 2:1 margin (63-35, D+28)"; yet to vote: "split 46 - 47 (D+1)". The press release says those still to vote were split 47–46.
• Enthusiasm: "80% of Democrats call themselves extremely motivated to vote, versus 67% of Republicans – a 13-point motivation delta." "Motivation falls to 47% among voters under 30 and 37% among swing voters."
• The ballot gap: "The difference is Trump." His net favorable was "12 points lower than in Oct 2024." Definite Republicans fell "from 41% in 2024 to 34%."
• Republicans needed Election Day: "Nearly half of Republican voters (48%) plan to vote on November 3rd, compared with 29% of Democratic voters." Election Day voters still to vote favored Republicans by 15.
• Late movement: "Swing voters are back with Republicans. Among voters who haven't cast a ballot, soft Republicans lead soft Democrats R+13." Press release: "Even those who decided who they would support in the last week favor Republicans by 10 points (51% to 41%)."
• Trust: "Cost of living moved the most, from a 1-point Republican edge to a 12-point Democratic lead (48-36). Immigration held at R+12 to R+11." Economy moved "from R+3 to D+6." Foreign affairs "from R+1 to D+5."
• Blame if Republicans lost Congress: "Trump takes the most blame (35%), followed by high prices and the cost of living (20%)."
• After a loss, "Nearly half of voters (47%) say Trump should have little or no influence over the party."
• Iran items: see section 3.
• Accuracy grade. Middle of the pack nationally (−4.6). Its finding that late deciders leaned Republican (51% to 41%) roughly fits the state exit polls. There, voters who decided in the last week leaned Republican in five of six Senate races with a usable split (Texas and Michigan 46-48, Ohio 42-49, Florida 43-54, Iowa 32-52; North Carolina went 57-40 for Cooper). But they were only about 6% to 12% of voters, and those are Senate-race exit polls, not Cygnal's national sample.
• After the election. Congratulation posts to winning clients and a CEO clip on AI and jobs. No post-election survey or accuracy statement. (Cygnal on X)
• Saved: Cygnal-National-Late-Oct26-NVT-Deck-Public.pdf. Sources: Cygnal deck; Cygnal press release, Oct. 27.

4.11 Ipsos

• FINAL PRE-ELECTION POLL. D+8.0; field ended Nov. 2; likely voters. AtlasIntel tied #16 (5.8). Morris #16 (−5.2, 4 polls).
• Alec Tyson: polls did well showing "the arc of the election" and "documenting the erosion of Trump's approval." "It's worth noting that the average of the generic ballot, even going into election night, was saying this has a chance to be a historic night." (The Hill)

4.12 Third tier: final national number only (D+7 to D+8)

All are FINAL PRE-ELECTION POLLS.

4.13 Bottom tier

Rasmussen Reports after the vote (POST-ELECTION SURVEY). "Election 2026: 38% Rate Media Coverage OK" (Nov. 4) and "Most Give Congress 'Poor' Rating" (Nov. 5). Daily Trump approval at 42% to 43% (a likely-voter tracking poll; it runs well above the exit poll's 37%). No accuracy review. Many of its Nov. 3–5 posts are about election-administration claims and are left out here. Sources: X; X; X.

4.14 Echelon Insights / Patrick Ruffini (Republican)

Who. A Republican polling firm. Co-founder Patrick Ruffini writes the newsletter "The Intersection." He was the most-quoted pollster after the election. No pollster named "Patrick Griffin" turned up in any search; the notes assume Ruffini was meant.

Accuracy grade. Near the bottom nationally: D+5 (AtlasIntel tied #29, error 8.8; Morris #20, −9.0 over 2 polls). Miss vs. D+13: −8.0. Ruffini's own turnout estimate (about 70% of Trump 2024 voters turned out, against 90% of Harris voters) may help explain why a D+5 read was too low. That link is our reading; he has not said it about his own poll.

• FINAL PRE-ELECTION POLL: October 2026 Verified Voter Omnibus. Oct. 15–19; 1,010 voters in the likely electorate; ±3.6. Generic ballot: "DEFINITELY/PROBABLY/LEAN THE DEMOCRATIC CANDIDATE 50%" vs. Republican 45%. Other findings (read from the saved file): right direction 30%, wrong track 63%; Trump job approval 39% approve, 60% disapprove; biggest issue (first choice) cost of living 33%, political corruption 13%, jobs and the economy 12%, the state of democracy 10%, immigration 7%, healthcare 5%, national security 4%. Party trust: Republicans led on immigration and border security (46-40) and crime (43-38). Democrats led on foreign policy (43-39), jobs and the economy (45-37), democracy (46-35), inflation and the cost of living (44-33), healthcare (50-32), AI (36-27) and keeping America out of wars (47-25). Favorability: Trump 41% favorable, 58% unfavorable; JD Vance 43/50; Ken Paxton 20/27 (36% never heard of him); James Talarico 27/24; Jon Ossoff 23/17. Iran items: section 3.
• Echelon's Election Day posts used this poll: 66% called the election "an existential decision" versus 23% "about differences in policy ideas"; 49% named "Economy and cost of living" as top of mind (57% of voters under 50, 43% of voters 50+).
• Saved: October 2026 - Voter Omnibus Topline - Updated External.pdf; HWORJS5WIAMj3hJ.jpg; HWOuCYLXcAEA1sn.jpg; HWOudozXcAEX2wA.jpg. Sources: Echelon PDF; Echelon on X; X.
• PRE-ELECTION SURVEY (issues): September Trump-regretter question. Sept. 17–21; 1,002 registered voters in the likely electorate; ±3.6. Asked of 112 Trump 2024 voters who now disapprove: economy and cost of living 42%; foreign policy and wars, including Iran, 15%; broken campaign promises 7%; leadership and direction 6%; personal character flaws 3%; self-interest over country 3%; tariffs and trade 3%; other 2%. Pull quote: "I can't afford to live." Ruffini's caption: "Not that complicated." Saved: PatrickRuffini-2117894200029307207-1.jpg. Source: Ruffini on X.
• Ruffini's post-election analysis (readings of the exit poll and vote counts, not new polls).
• Exit-poll swings from the 2024 presidential exit poll to the 2026 House exit poll (preliminary, 9:15 p.m. ET Nov. 3): ages 18–29 D+35; Asian D+25; Latino men D+24; Latino D+20; suburban D+20; independents D+13; 65+ R+3; Black women R+1. Title: "Democrats gained almost everywhere. Young voters moved most." (X)
• Among groups that moved right in 2024, net versus 2020: Latino men R+33 then D+24; Latinos R+27 then D+20; voters 18–29 R+13 then D+35. (X; 3.79 million views)
• "Latinos are actually better than they were in 2020 even on this horrendous night for the GOP, suggesting many of these gains may be durable." This conflicts with Cohn's NYT data, which has Hispanics past their 2020 level. (X)
• ESTIMATE/MODEL: "Just going off county turnout vs. 2024 points to 70% of Trump 2024 voters voting and 90% of Harris 2024 voters voting." (X)
• Issues: "Your personal pet issue - the war, data centers, tariffs - mattered, but they only mattered in how they laddered up to broader arguments about affordability and corruption. The war = gas prices / Data centers = rising utility rates, corrupt deals / Tariffs = prices." (X)
• "Turnout and persuasion are the same thing. If they're sitting on the couch, you haven't persuaded them." (X)
• Senate polls vs. NYT live estimates (election night, partial counts): polls understated Democrats in Texas by 5.6, Iowa 3.6, Mississippi 3.5, Ohio 2.8, Minnesota 2.0, Kansas 0.7; overstated them in Maine (0.1), Michigan (0.7), Louisiana (1.8) and Nebraska (2.2). (X)
• Texas page: "The 17 counties where Hispanic voters are at least 70% of the electorate swung 43 points from 2024." "No county moved toward the Republican." (The Intersection)
• No Echelon post-election survey yet. Its website and Substack (last post Oct. 30) show none.
• Saved: echelon-insights folder (12 images and the October PDF).

4.15 Not in either national ranking

Marquette Law School Poll. FINAL PRE-ELECTION POLL, national: D+13 (field dates and population not in our research). FINAL PRE-ELECTION POLL, Wisconsin governor: Crowley +7 among likely voters vs. an unofficial result of +8.0. Marquette: "The national MULawPoll congressional ballot had D+13, which people also said was high. Looks like +13 to +14." On D+13 it would have no miss; against the incomplete D+11.4 count it ran 1.6 high. (Marquette on X)

Big Data Poll. FINAL PRE-ELECTION POLL. "Democrats lead Republicans 52.3% to 40.0" (D+12.3); Oct. 27–30; 3,418 registered and 3,214 likely voters; published Oct. 31. Nothing post-election. (Big Data Poll)

Trafalgar Group (Republican). All FINAL PRE-ELECTION POLLS of likely general-election voters, ±2.9 (read from the saved files).

Response rates were 4.4% to 5.4%. Flag: in its South Carolina file, voters aged 18–29 split exactly 93.0 to 7.0 in both the Senate and governor races, which suggests a very small subgroup. No post-election review found. Misses are our arithmetic on preliminary counts. Saved: ME-Midterms-Report-1025.pdf, MI-Gen-Sen-Poll-Report-1101.pdf, OH-Midterms-Report-1102.pdf, SC-Midterms-Survey-Report-1101.pdf, Screenshot-2026-11-01-at-9.38.51-PM.png. Sources: TX; ME; MI; OH; SC.

Suffolk University. No final number in our research. David Paleologos: "We were low on the margins of victory." "We were all undercounting Democrats." "I think to be fair going forward, it can't be a one-sided question anymore." He noted the same undercount in the 2025 New Jersey and Virginia governor races. (Fox News)

InsiderAdvantage (Matt Towery). No final number in our research. Election Day: "I am told that GA same day vote may go well beyond the projected 1.4 million. If so this is huge positive news for the GOP there. AND FL same day is well outpacing projections, again a shocker to the positive for the GOP." Georgia went to Ossoff by 18. No review found. (X)

McLaughlin & Associates (Republican; Trump's 2024 pollster). No poll published. On Newsmax, Nov. 4: "We lost independents 56%-39%. We did 10 points worse with Hispanics." "We got single digits with African American men, not the 21% we got [previously]." "If you don't have a message and you don't have a strategy, this is what happens." He did not say where his numbers came from; the exit poll has independents at 58-38. (Yahoo/Newsmax)

Quantus Insights. Published a final model (no numbers in our research) and a vote-count post, not a poll: "Starr +72, Zapata +63, Maverick +60, Webb +49, Hidalgo +49, Cameron +44. The 2024 Hispanic shift to Trump fully reversed and then some." (X)

Data for Progress (Democratic) and Navigator Research (Democratic): no national final in the rankings. Their material is in section 6.

5. The exit poll: the 2026 network Voter Poll (SSRS)

5a. What it is and which version this uses

• One survey, many names. In 2026 the old Edison exit poll and AP VoteCast were replaced by one survey run by SSRS. CNN calls it the "CNN Voter Poll," AP the "AP Voter Poll," Fox the "Fox News Voter Poll," NBC the "NBC News Exit Poll," and Morris cites the "MS NOW Voter Poll." When outlets differ, they are usually reading the same survey at different hours.
• How it works. "Nearly 80,000 voters" from a pre-election survey (Oct. 19 to Nov. 3) plus in-person exit interviews in 15 states ("Roughly 22,000 voters are expected"): Alaska, Florida, Georgia, Iowa, Maine, Michigan, Minnesota, Nebraska, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Texas and Wisconsin. Every respondent was matched to the voter file. The pre-election part drew on the SSRS Opinion Panel (probability-based) and the Verasight panel (opt-in). Margin of error "plus or minus 0.6 percentage points" (CNN, Nov. 3) or "plus or minus 0.5" (CNN, Nov. 5; Fox). AP: "Topline vote shares are not reported." Sources: CNN; AP methods.
• Version used here, unless marked. CNN's national House file, stamped Nov. 4, 16:30 UTC (11:30 a.m. ET), "stage 8," 79,668 respondents, downloaded Nov. 5 at about 11:03 a.m. Pacific. State files are stamped Nov. 4, 12:07 to 3:56 p.m. ET, "stage 9" (Arizona: 2:36 a.m. ET, "stage 7"; Nevada: "stage 8"). Fox's results table, downloaded Nov. 5 at about 11:08 a.m. Pacific, matched. All can still be reweighted.
• Label for every number in this section: EXIT POLL (SSRS Voter Poll). Outlet and time are noted where they differ from the CNN file.
• Saved: cnn/api folder; 2026-HG-US.json; voter-poll-results.html; 2026-Voter-Poll-Methods-Statement.pdf (only a link to a web page).
• Our arithmetic, not a published number: the survey's own groups add up to about a 12-point Democratic House win (about 55% to 43%). Use the real count for the topline.

5b. Vote by group

2026 share of voters and House vote (D-R), with the swing from the 2024 House exit poll. Swing is our arithmetic. Source: CNN 2026 file and CNN 2024 House file.

Caution on shares. The 65+ share jumped to 36% (from 28%) and college graduates to 47% (from 43%). The new method is mostly voter-file-matched pre-election interviews, which may explain part of this. Voter files also contradict the 9% Black share (see the combined report). Trust the vote splits more than the shares.

What the outlets wrote: - NPR (Nov. 4, 12:43 a.m. ET, updated 6:41 p.m. ET, against the 2024 House exit poll): "Latinos swung 21 points toward Democratic House candidates." Young voters: "Democrats won this group of young voters by 48 points, a 34-point swing." "Democrats won men by 2 points, a 16-point shift from 2024." Republicans "only won 38% of independents, their worst mark since at least 1998." (NPR) - AP (Nov. 3, 10:04 p.m. ET): "about 7 in 10 voters under 30 nationally say they supported the Democratic candidate." "Only about one-third of Hispanic voters say they backed the Republican ... while roughly 4 in 10 voted for Trump two years ago." (AP via PBS) - Fox (Nov. 4, 12:10 p.m. ET, against 2024 AP VoteCast): "Independents backed Democratic candidates by 22 points, a 16-point shift from 2024, when they backed Harris by just 6 points." "Hispanic voters ... backed Democratic House candidates by 27 points, after breaking for Harris by just 12 points." "Young men made a complete reversal ... going from a 15-point margin for Trump to a 33-point margin for the Democratic House candidates, a 48-point swing." Rural voters: Trump +28 to R+18. Suburban women: Harris +14 to D+24. (Fox) - NBC (Nov. 4): Latinos 62% D / 35% R; men under 25 from Trump +1 (49–48) in 2024 to Democrats 68–28; Black men 13% for GOP House candidates. (NBC) - CNN: "The double-digit advantage that GOP candidates held with men in 2022 was wiped out this year." (CNN)

5c. Crossovers and the 2024 vote

The real 2024 result was Trump 49.8%, Harris 48.3% (FEC; saved 2024presgeresults.pdf). So 2026 voters leaned about 6.5 points more toward Harris than the 2024 electorate. That gap can come from Trump voters staying home, people misreporting their 2024 vote, or weighting. A survey can't separate them.

• CNN (Nov. 5): "Nationally, about 7% of self-reported Trump voters who turned out in this year's election said they were backing a Democratic congressional candidate." "just 2% [of Harris voters] said they'd voted for a GOP candidate." In Texas, Ohio, Iowa, Kansas and South Carolina, "roughly one-tenth of those Trump voters backed the Democratic candidate." "20% of 2024 Trump voters now say they disapprove of him." "among the midterm voters who backed him in 2024, only about one-third say they viewed their House vote this year as a message of support for the president." (CNN)
• NBC's earlier read of the same survey: 6% of Trump voters went Democratic and 3% of Harris voters went Republican.
• Crossover by group (Trump voters who voted Democratic): ages 18–29 18%; Latino 13%; Black 27% (small group); white no degree 5%; 65+ 4%. Non-MAGA Republicans gave Democrats 15%; MAGA Republicans 1%.
• Fox: "54% identified as MAGA, and 99% of them voted Republican. Among non-MAGA Republicans, 85% stayed with the party." (Fox counts leaners.)
• MAGA share of all voters, counting leaners: 36% in 2024 (AP VoteCast via Fox) and 23% in 2026. Different surveys, so treat the drop as a sign, not a measurement.

5d. Issues

Top issue (one choice; abortion was not offered in 2026)

Fox by party: "Democrats: democracy 48%, economy 36%. Independents: economy 39%, democracy 34%. Republicans: economy 41%, immigration 27%, democracy 16%."

How important each issue was (also reported by ABC, Nov. 4)

Economy and prices

The cleanest finding: voters who called the economy bad went Trump 70-28 in 2024 and Democratic 70-28 in 2026. The mood barely changed. The blame did.

Outlet versions: NBC said "62%" blamed Trump for high prices and trusted Democrats over Republicans on the economy 43% to 37% (CNN's later file: 44-37). AP: "about two-thirds of voters nationwide said they had made the economy worse. And out of this group, about 8 in 10 said they backed the Democratic candidate." About 9 in 10 called the cost of living a crucial or major factor, and "About 2 in 10 Texas voters said inflation had imposed 'severe' hardship," three-quarters of whom backed Talarico. (NBC; AP via PBS; AP via KSAT)

Iran, immigration and other issues

Which party is better on...

5e. Trump approval and the national mood

Fox: "Job approval was 38%-62%, a net rating of -24 ... At the 2018 midterm point, his approval was 44%-56%." Conflict: Fox puts Trump's 2018 approval at 44%; NBC and the CBS card put it at 45%. NBC called the 2026 result "a shift in mood, not ideology": liberals 25% (23% in 2024), conservatives 34% (35%). CNN's later file has 26% and 33%.

Saved charts: dcg1114-2117745361133461941-1.jpg (dcg1114 on X); williamjordann-2118316526843613625-1.jpg (Jordan on X); williamjordann-2117877228193362235-1.jpg; williamjordann-2118074132562862398-1.jpg.

5f. State exit polls for the key Senate races

From the CNN state files (Senate races, plus governor races in states with no Senate race). Iran items are in section 3b.

When voters decided (asked only in Senate races): voters who decided in the last week, counting the last few days, were 6% to 12% of the electorate. Where the sample allowed a split, they mostly leaned Republican: Texas and Michigan 46-48, Ohio 42-49, Florida 43-54, Iowa 32-52; North Carolina's went 57-40 for Cooper. Democrats built their leads among voters who decided before October. About 9 in 10 voters had decided before the last week, so this looks like a judgment voters had already made, not a late collapse.

Candidate questions (AP): "About two-thirds of Texas Senate voters said he [Paxton] lacked the character to be a senator." In South Carolina, "About 6 in 10 voters ... said she lacked the experience to serve in the Senate." In North Carolina, "about 6 in 10 voters said ... Michael Whatley, was 'too supportive' of Trump." NBC: only 34% viewed Paxton positively; Ossoff 59%, Talarico 52%, Turek 49% favorable. CBS on Maine: "Most Maine voters (69%) feel [Trump's tariffs] have harmed their local economy" (CNN's later Maine file: 71%).

Sources: CNN state files (pattern 2026-SG-XX, 2026-SP-XX, 2026-GG-XX); AP via PBS; CBS. Kansas and South Carolina have about 1,000 respondents, and they and Arizona had no in-person interviews. Iowa, Kansas, Maine and South Carolina had no 2024 state exit poll to compare against.

5g. Where the outlets' versions differ

The 2024 baseline changes the swing. The 2024 Edison exit poll and 2024 AP VoteCast described Trump's coalition differently. Pick one and say which.

Also note the swings in section 5b compare with the 2024 House exit poll (for example, ages 18-29 +34). The table above and Ruffini's chart both compare with the 2024 presidential exit poll. Ruffini's chart shows a 35-point swing for ages 18-29, not 37, most likely because it used preliminary election-night numbers.

Other Latino estimates for comparison. Exit poll 64-34 (D+30). Fox and MS NOW D+27. UnidosUS 66-31 (D+35). AtlasIntel 61.6-37.8 (D+23.8). NYT Voter Trends 69-29 (D+40). Publish a range, not one number.

6. Other surveys, focus groups and estimates

6a. Surveys

UnidosUS / BSP Research Latino voter poll. POST-ELECTION SURVEY (advocacy-sponsored). - BSP Research ran it with the African American Research Collaborative, Harvard and Penn State. 10,141 voters in all, including 3,014 Latino voters (±1.8%). Respondents had already voted or said they were certain to vote on Nov. 3; fieldwork ended the morning of Election Day. Live phone, text invitations, email and online panels, in six languages including English and Spanish. - Flags: the dates differ by document ("Oct 20 – Nov 3" in the deck; "October 19-November 2" in the crosstabs; Oct. 19 to Nov. 3 in the methods note). The deck's footers say "N=3,050" while its survey page says N=3,014. A few numbers differ by a point between the deck and the crosstabs (for example, Iran approval disapprove 69 vs. 70). - Vote: Latino House "66" D vs. "31" R (D+35). Senate 63-33. Governor 66-30. House by state: Arizona 73-25, California 70-27, Florida 58-40, Michigan 69-30, Texas 65-34. Men 63-35, women 69-28. Recalled 2024 vote: Harris 56, Trump 34. Latino Trump 2024 voters who voted Democratic for House: 17% (men 15%, Latinas 19%; Texas 17%, Florida 13%, Michigan 28%). Harris voters who voted Republican: 3%. Texas Senate: Talarico 65, Paxton 31; Rio Grande Valley 74-24; Latino Trump voters 19% Talarico, 75% Paxton. Florida Senate: 54-42. - Mood and issues: wrong direction 63%, right direction 31%. Top-three issues: cost of living/inflation 58, jobs and economy 35, cost of gas 30, health care costs 29, affordable housing 29, immigration reform/ICE 27, stopping the Trump agenda 20, corruption 10, war with Iran/foreign policy 5. Cost-of-living worries: food 54, housing 41, gasoline 38, utilities 32. Half "have nothing left over or are struggling." 43% expect Trump's economic policies to leave them worse off. - Views: Trump favorable 32, unfavorable 64. Trump's handling: economy 32/64 (net −32); the war with Iran 25/69 (net −44); immigration enforcement and ICE 30/68 (net −38); inflation and cost of living 26/71 (net −45). Democrats care "a great deal" about Latinos: 63%. Republicans: 32% care, 37% don't care too much, 32% hostile. 63% worry their community "could be targeted by government policies." Data centers near their community: support 36, oppose 53. - Latino Trump voters who switched (n=131, very small), top issues: cost of living 57, jobs and economy 43, health care costs 34, immigration reform/ICE 32, cost of gas 30, "stopping the Trump agenda" 27. - Saved: unidosus_2026_midterm_election_voter_poll_deck.pdf; unidosus-national-crosstabs-2026-midterm-voter-poll.pdf; 2026_Midterm_Election_Voter_Poll_Methodology.pdf. Sources: deck; crosstabs; methodology.

CIRCLE (Tufts) youth vote estimate. POST-ELECTION SURVEY (published Nov. 4; field dates not in our research; data from the NORC Civic Health & Elections Project). "11% of all votes in the 2026 midterms were cast by youth, which is slightly lower than the 12% youth share" in 2022. "66% of young people ages 18-29 voted for a Democratic candidate for the U.S. House of Representatives," against 31% Republican. 70% of young women and 59% of young men voted Democratic. White youth went 61-36 Democratic. Rural youth favored Republicans 50% to 44%. Its youth turnout estimate was not out yet. Conflict: the exit poll has under-30 voters at 73-25. (CIRCLE; CIRCLE demographics)

Fox News / NORC AmeriSpeak Study. PRE-ELECTION SURVEY (issues), Nov. 1–3, "approximately 3,000 registered voters" from NORC's probability-based panel, online and by phone, English and Spanish. Iran findings in section 3. Flag: the notes give 3,464; the saved Fox page says about 3,000.

Data for Progress (Democratic). PRE-ELECTION SURVEY (issues), Oct. 31–Nov. 3, 1,188 likely voters. "How much, if anything, have you heard about each of the following in the news recently?" Heard "a lot": cost of living 66%, the war with Iran 57%, AI 55%, data centers 52%, immigration and deportations 48%, corruption allegations in the Trump administration 44%, Flock Safety cameras 34%, protests in France 32%, Medicaid cuts 31%, the World Series 26%, the Fed raising rates 19%. No post-election release; its blog's newest item was Oct. 30. Saved: HWOV6zNXUAEZ_B0.jpg. (DFP on X)

Navigator Research (Democratic). PRE-ELECTION SURVEY (issues), recirculated after the vote. No new post-election survey as of Nov. 5. Its posts reused October data: "Trump approval is tanking (net -26) and Navigator tracking shows it's at an all-time low!" and "One-in-four Americans who voted for Trump in 2024 now regret their vote" (reports dated Oct. 8 and Oct. 7). A Nov. 5 graphic says "73% of Americans view the economy negatively" (survey date not shown). Saved: navigator-research folder. (X; X; X; Navigator report, Oct. 7)

Gallup. PRE-ELECTION SURVEY (issues). Newest items were pre-election: "The Issues Voters Say Matter Most This Midterm" (Oct. 29) and "Voter Engagement High for Midterms." Their numbers are not in our research. (Gallup)

6b. Focus groups

Navigator Research / GBAO. FOCUS GROUP (Trump regretters; date not given). Margie Omero on NPR listed "Costs / Iran War / Broken Promises / Corruption." Will Jordan cited Navigator's Gen Z open-ended answers: "The three ee's: Epstein. Economy. Iran." No numbers. (Navigator on X; Jordan on X)

No other focus group appears in our research.

6c. Estimates and models (not polls)

Why pollsters say they missed. Chaz Nuttycombe (State Navigate) said many pollsters weighted to a 2022-like electorate and cited recall bias: "You had a whole bunch of people who were lying, who are definitely Trump voters who said that they weren't." Axios: "A Pew analysis found that nearly 70% of national polls used such political weighting in 2024." Cohn: weighting to the 2024 electorate "obviously wasn't a solution." Jain: "herding." Silver: unpopular-Trump voters may not admit past Trump support, which skews polls weighted to past vote. (The Hill; Axios)

7. Who hasn't published yet

Nothing post-election had been found from these as of Nov. 5. The notes say most firms historically publish post-election surveys one to three weeks after the election (for example, Navigator's 2024 release was Nov. 20, 2024, per search results). Re-check around Nov. 12 to 25.

Our own gaps, not the pollsters': AtlasIntel's national-poll chart numbers (issues, Trump's handling) were not extracted from its PDF. The NYT Voter Trends page, Ruffini's paid newsletter, Silver Bulletin's paid sections, and Daily Signal's Nov. 5 piece were not opened. SSRS's own 2026 methods page was not opened. X searches late on Nov. 5 returned errors, so later posts from Neil Newhouse, Glen Bolger, Brent Buchanan, Ryan Munce, Frank Luntz, Gallup and Pew were not checked. Results for the New Hampshire and Minnesota Senate races were not in our research.