US midterms 2026: Latest polls and updated House and Senate outlook

Votendi election analysis — updated October 13, 2026

With three weeks remaining before the November 3 midterm elections, national congressional polling, a series of close Senate surveys and significant changes in independent House race ratings have produced a more detailed picture of the electoral map. The central distinction is between what voters told pollsters, how outside election analysts classify individual races, and what the arithmetic of a possible congressional majority actually requires.

Votendi’s U.S. election database contained 183 published polls at the time of this review, with the most recent poll publication dated October 12. Its national generic congressional ballot average stood at 49.2% Democratic and 41.1% Republican — a Democratic margin of 8.1 percentage points, calculated from 67 included polls. This is a polling measurement, not a nationwide seat estimate: the House is elected district by district, Senate elections occur in individual states, and the two chambers have different electoral maps.

The latest national surveys reinforce a substantial Democratic lead, while close state and district polls leave the balance of Congress dependent on a smaller group of competitive contests. This update reviews the polling evidence, identifies the races that warrant reassessment and sets out Votendi’s conditional central seat scenario.

Votendi projection: Democrats favored for the House, Senate majority remains close

Votendi’s October 13 central editorial scenario is 226 Democratic seats to 209 Republican seats in the House, and 51 Democratic-aligned seats to 49 Republican seats in the Senate. This is an independent, judgment-based conditional projection using the polls in our database and an evaluation of individual contests. It is not the output of a calibrated statistical forecasting model, and these numbers should not be interpreted as implied win probabilities.

The assessment combines national polling with state and district surveys, recent race-rating updates and the limitations of the available evidence. The Cook Political Report House ratings, updated October 9, and its Senate ratings, updated October 6, provide external context for the electoral map.

Five key races in the updated outlook

These provisional classifications reflect the polling evidence reviewed on October 13. Maine’s surveys point consistently in the same direction; Ohio remains a lower-confidence call because its polls conflict. TX-15, KY-06 and WA-03 require particular care over sponsorship, fieldwork dates and the limited number of independent surveys. The article does not change Election Pulse poll records or live race-status fields.

How our projected majorities are calculated

House: Votendi puts 213 seats in Democratic-favored categories, 200 in Republican-favored categories, and 22 in toss-up. Our central scenario assigns 13 of those 22 toss-ups to Democrats and nine to Republicans, producing 226–209. With the same non-toss-up calls, Democrats need only five toss-ups to pass the 218-seat majority line. Moving five or more races against this scenario would materially change the projected margin.

Senate: our rated balance is 49 Democratic-aligned, 46 Republican, and five toss-ups: Alaska, Iowa, Kansas, Michigan and Texas. Assigning two of those five to the Democratic-aligned bloc produces 51–49. Michigan and Alaska are the two provisional Democratic allocations in this central scenario, informed by narrow recent leads, not a claim that either is secure. Iowa, Kansas and Texas remain Republican allocations in that scenario, despite demonstrably competitive surveys. Other two-of-five combinations could produce the same balance.

What does the latest evidence show? Maine and TX-15 show Democratic advantages in the reviewed surveys, while Ohio’s conflicting results make its Democratic lean tentative. WA-03 remains closely balanced, and the evidence in KY-06 warrants a competitive assessment with caution about partisan polling. These are current editorial assessments, not proof that all five races shifted after the previous update. The national surveys reinforce the broad Democratic advantage rather than establish a sudden new surge.

The national picture: a substantial generic-ballot gap, but not 435 identical races

The two latest national surveys recorded in Election Pulse are Morning Consult, with Democrats at 49.7% and Republicans at 43.1% among its reported survey population over October 6–12 (n=17,654), and Cygnal, with 51% to 42% over October 7–10 (n=1,500). These give respective Democratic margins of 6.6 and 9.0 points. They should not be mechanically pooled as if sampling methods, weighting and population definitions were identical.

The RealClearPolling average is included as an external comparison, not an additional survey in Votendi’s database. Different sampling windows and aggregation rules explain why two averages need not agree exactly. More importantly, an eight-point national margin does not translate directly into an eight-point shift in every congressional district or a fixed number of seats. Turnout, district lines, incumbency, local candidates and the distribution of voters all matter.

What actually changed: the House ratings on October 9

The Cook Political Report changed 18 House race ratings on October 9, with every change moving in the Democratic direction under Cook’s classification system. These were documented changes to an external rating framework, not 18 new election results or 18 new polling releases.

These changes are also recorded in the dated 270toWin Cook ratings archive. They describe changing assessments of competitiveness. They do not establish how a district will vote on November 3.

The Senate changes were narrower — but strategically important

Cook’s October 6 Senate update recorded three changes: Kansas from Lean Republican to Toss-up; New Hampshire from Toss-up to Lean Democratic; and North Carolina from Lean Democratic to Likely Democratic. In a Senate split across 100 seats, individual states can materially change the majority calculation. The significance of the Kansas change is not that a winner has been determined; it is that a state previously placed in a less competitive category has moved into Cook’s most competitive bucket.

The latest state polling is not uniform. Some surveys show multi-point margins while others find effectively level contests. The following table reports selected individual polls already recorded in Votendi; it does not turn those surveys into outcome probabilities.

The fieldwork end date is listed because it is not always the date a survey is published or added to an aggregator. An October 13 database addition of a September poll would be a newly tracked record, not fresh October interviewing.

Maine, Ohio and North Carolina: separate survey patterns

In Maine, the Washington Post/Schar survey and Rasmussen show different-sized margins, but the same order of reported support. An additional Marist survey conducted October 1–4 among 1,505 registered voters reported Jackson 51%, Collins 47%. The Marist population is registered voters, so its margin should not be treated as directly interchangeable with a likely-voter sample. The consistent direction across methods is notable; the differences in sample composition still matter.

Ohio is a clearer reminder why individual polls should not be reported as election results. CNN/SSRS reported 49–43 for Sherrod Brown over Jon Husted, while co/efficient subsequently recorded 43–44. Different samples, modes and undecided-voter treatment may contribute to the disagreement. Any Ohio analysis should disclose both rather than selecting the preferred result.

In North Carolina, the latest Votendi records place Roy Cooper above Michael Whatley in separate October surveys. Reuters also describes a marked spending and fundraising disparity between the campaigns. Even where several polls point in one direction, a survey lead is a measurement on the date of fieldwork, not a certified seat. See Reuters’ October 12 state-by-state reporting.

Michigan and Alaska: caution about methodology and electoral rules

Michigan’s October 10 InsiderAdvantage entry shows El-Sayed 48%, Rogers 46%. The Votendi database also contains a separate Wedgewood Polls entry (poll ID 328, October 8–9 fieldwork, 607 respondents) reporting 53–47. That Wedgewood entry is still marked “published” in Election Pulse as of this review. Its online-panel recruitment, representativeness and treatment of undecided voters have not been sufficiently verified for a confident methodological comparison. Votendi therefore flags the survey in editorial interpretation rather than treating its larger margin as corroboration of the better-documented, closer result. An administrative exclusion has been requested for later review but has not yet been applied.

Alaska requires a different caveat. The state uses ranked-choice voting and has multiple candidates on the ballot, including two candidates named Dan Sullivan. A reported 50–48 head-to-head or final-round preference should not be conflated with the first-choice vote or with an independent second poll. The distinction is especially important when charting changes over time. Associated Press reporting on October 13 outlines the ballot and the campaign context.

Texas, Kansas and Iowa: why narrowly separated surveys matter

In Texas, Trafalgar recorded Talarico 47.1% and Paxton 43.6%, while the UMass Lowell/YouGov entry in Votendi shows a 45–45 tie. Both results belong in the story. It would be unsound to assume that a three-and-a-half-point gap in one organization’s poll has erased the close result in another.

In Kansas, InsiderAdvantage’s 48–47 survey falls within an extremely narrow interval, consistent with the broader reassessment of the race’s competitiveness. In Iowa, both the Emerson survey (Hinson 47, Turek 45) and CNN/SSRS (45–43) record small Republican margins. Single-digit differences across close surveys are vulnerable to sampling error, turnout-model error and late shifts in who answers polls.

House districts in Votendi’s tracking database: fresh records versus fresh fieldwork

Two of the newest entries added to Votendi illustrate the point. In NY-17, a late-September Emerson poll records Cait Conley 48%, Michael Lawler 46%. An earlier survey in the database showed Conley 45%, Lawler 50%. Those measurements indicate differing survey snapshots, not a precisely measured seven-point change within one continuous panel. Cook’s October 9 map still classifies NY-17 as a toss-up.

In TX-15, a September 24–28 Normington, Petts & Associates survey reports Bobby Pulido 54%, Monica De La Cruz 41%. Its commissioning organization was Democratic-aligned House Majority PAC, a material disclosure when weighing the result. A separate Texas Southern University survey registered 50–45. These surveys are relevant evidence, but the sponsor, small sample of the later poll and the limited number of independent surveys require additional care. Cook currently places TX-15 in the toss-up group. Axios’ South Texas reporting discusses the broader electoral context.

Meanwhile, WA-03 demonstrates why a race-rating change is not the same thing as a new poll: Cook moved it from toss-up to Lean Democratic on October 9 even though a previously released DHM Research poll showed a near tie, Braun 48% and Gluesenkamp Perez 47%. Analysts incorporate district-level structure, previous results, spending and campaign information beyond any single questionnaire.

Why the final three weeks may differ from the polling snapshot

Reuters’ October 12 analysis points to several relevant sources of uncertainty: geographic concentration of competitive House seats, potentially imperfect turnout models, unequal campaign resources, redistricting and the possibility of systematic polling error. The news agency estimates that roughly 40–50 House races are genuinely competitive, far fewer than the chamber’s full 435-seat membership.

That is why Votendi treats national averages and local surveys as separate types of evidence. A national D+8 survey can coexist with a state race within a single point. A polling organization’s large sample reduces some random uncertainty but does not cure every possible nonresponse or likely-voter modeling bias. A party-sponsored district survey may be useful but deserves visible labeling rather than automatic parity with a transparent independent field study.

Our next data checks will focus on newly published fieldwork, whether competing pollsters confirm narrow local margins, and whether the exclusion of the methodologically flagged Michigan Wedgewood entry changes any displayed numerical average. The administrative status of that poll has not been changed as part of this article.

Methodology and primary sources

Database cutoff: Votendi Election Pulse checked October 13, 2026; most recent national publication October 12. The Votendi generic ballot average includes the site’s weighted published records and can differ from other poll aggregators. State tables are individual surveys, not forecasts; within-state entries were selected for recency and relevance, not to engineer a particular party result. Published percentages are retained in the pollster’s reported form without redistributing undecided voters. Cook rating counts are externally sourced and clearly labeled. Votendi’s independent ratings and 226–209 House / 51–49 Senate central scenarios are editorial judgments based on the reviewed polling, not probability-calibrated statistical forecasts; no formal model fit, district-level simulation or pollster-house-effect correction has been claimed.

Sources: Votendi U.S. election hub; Cook Political Report House ratings; Cook Political Report Senate ratings; RealClearPolling generic ballot; Marist Maine survey; Reuters Senate reporting, Oct. 12; Reuters polling-risk analysis, Oct. 12; OPB Washington-03 poll; AP Alaska reporting, Oct. 13.