2026 Senate battlegrounds: eight races to watch

Collins's history of beating expectations is the starting point

The central evidence is not simply that Collins has won in a Democratic-leaning presidential state. It is that several earlier Senate surveys understated her eventual margin. In 2020, every one of the 14 head-to-head polls in RealClearPolling’s archive had Gideon ahead, with not a single listed survey showing Collins in the lead. The final Emerson poll had Gideon up six points; Collins ultimately won by 8.6, a 14.6-point difference in the margin compared with that poll. In 2014, an August polling average placed her ahead by 29.2 points, while she eventually won by 37. She also won by 23 points in 2008, larger than a late Rasmussen survey's 16-point margin. Those comparisons cover different pollsters and dates, not a single consistent forecasting model, but they establish a history worth studying. Her Kavanaugh vote, however, is a concrete reason that older patterns may not transfer perfectly to 2026.

2020: the most striking polling miss

In 2020, every one of the 14 Collins–Gideon head-to-head surveys in RealClearPolling’s archived poll table showed Gideon leading. Not one of the listed polls showed Collins ahead. That is why the result was so striking: this was more than a miss by one pollster. Even the final Emerson survey placed Gideon at 48% and Collins at 42%, a six-point Democratic lead. Collins instead finished with approximately 51.0% to Gideon’s 42.4%, an 8.6-point Republican margin. That is a 14.6-point difference between the margin in that particular poll and the final result. The surveys varied in field dates, sampling and how they measured other candidates, so the magnitude of their errors differed. The claim about every survey applies specifically to the 14 published head-to-head polls displayed in RealClearPolling’s archive. But the reversal matters because Collins secured another term in a year when Democratic presidential nominee Joe Biden also won Maine by about 8.7 points. Collins continued to receive support outside the Republican presidential coalition.

The pattern extends to 2014 and 2008

The evidence did not begin in 2020. Maine Public reported that a RealClearPolitics average in August 2014 showed Collins leading Shenna Bellows by 29.2 percentage points; she ultimately won by approximately 37 points. In 2008, the final result gave Collins a roughly 23-point victory over Tom Allen, compared with a 16-point lead in Rasmussen’s November 1 survey. These figures compare snapshots from different points in the campaign, so they should not be treated as identical measures of polling error. Together, however, they show why analysts examining Maine pay attention to Collins’s capacity to outperform particular surveys and to win votes beyond a typical Republican nominee.

Why her crossover appeal matters

Collins has long presented herself as an independent-minded Republican, highlighting cross-party votes and work on behalf of Maine. Her electoral record suggests that at least some voters distinguish her from the national Republican ticket. Her 2020 victory is a particularly clear example: voters selected Biden for president statewide while also reelecting Collins to the Senate. The analytical challenge is identifying how much of that coalition is still available in 2026. Historical overperformance is relevant evidence, not a numerical correction that should automatically be added to a new polling average. Different opponents, turnout levels and campaign issues can change the electorate.

The Kavanaugh vote is a serious counterweight

In October 2018, Collins voted to confirm Brett Kavanaugh to the Supreme Court. Kavanaugh later joined the majority in the 2022 Dobbs decision overturning Roe v. Wade and ending the federal constitutional protection for abortion. Reporting by the Associated Press documents frustration among some Maine voters, including people who previously supported Collins because they expected her to protect abortion rights. Jackson has made the vote part of his criticism of her independence. Collins says she supports abortion rights, disagreed with the Dobbs ruling and does not regret supporting Kavanaugh; she also points to her support for other justices. The electoral effect cannot simply be assumed or assigned a percentage without reliable evidence, but the controversy provides a substantive reason to question whether every past Collins supporter will vote the same way again.

What today's polling says—and what it doesn't

An October 1–4 Marist survey of 1,505 Maine registered voters measured Jackson at 51% and Collins at 47%, with a reported margin of sampling error of plus or minus 3.8 percentage points. Interestingly, while more respondents in that sample said they would vote for Jackson, 57% said they expected Collins to win. A preference question and an expectation question are measuring different things; this gap does not by itself prove that Collins will repeat her historical overperformance. For the site’s analysis, current polling should remain visible and separate from judgments about historical polling errors.

How this history fits the published forecast

The central tension in Maine is straightforward: Collins has repeatedly performed better than certain historical surveys suggested, but her Kavanaugh vote and subsequent anger among abortion-rights supporters may make 2026 different. Will’s current rating and prediction remain whatever the live Senate map displays. A transparent written rationale should explain how much weight it assigns to Collins’s prior overperformance while considering her opponent, the current polling sample and possible changes in voter attitudes. New independent surveys, especially consistent likely-voter polls, would help show whether Jackson’s support persists or the measurements move. Neither the 2020 surprise nor the Kavanaugh controversy alone determines the 2026 result.

The partisan baseline

Texas has a long record of Republican statewide victories. That history matters because an unusually close Senate poll has to be understood against the much broader electorate that has voted Republican in recent elections. At the same time, a state's historical party alignment is not a fixed percentage: different candidates, issues and turnout levels can produce a result that differs from a past presidential race.

Why the surveys look so competitive

The University of Texas/Texas Politics Project survey reported James Talarico at 50% and Ken Paxton at 47%, while an October InsiderAdvantage poll listed Paxton at 49% and Talarico at 48%. Those findings are not identical, and their differences matter. Compare sampling populations, field dates and methodology before describing the trend. A two- or three-point margin can easily be smaller than the combined uncertainty surrounding an election estimate.

Candidate differences and campaign resources

Talarico and Paxton have distinct public records, campaign messages and fundraising operations. Their campaigns are fighting over economic conditions, national political identity and the candidates' personal and governing records. Heavy advertising is evidence that campaigns are investing in the race; it is not direct evidence that an advertisement has changed someone's vote. Suburban participation and whether occasional voters turn out are also important analytical questions.

Why a forecast is not just the average

Will's published prediction may place different weight on current candidate-specific polling and Texas's longer Republican voting history. A well-explained rating should make that tension explicit instead of treating either the historical margin or today's poll as decisive. For example, a narrow polling advantage can coexist with uncertainty about which voters will actually cast ballots. Any numerical probability on the live site is a forecast assumption, not a number measured by a survey.

What to watch next

Track whether Talarico and Paxton move in the same direction in several polls taken over comparable dates; monitor the share undecided, turnout screening and how results vary by pollster. Separate a genuine multi-poll trend from random sampling variation. An additional poll with a larger sample helps only if its design, transparency and voter population are appropriate.

Why this is an unusual rematch

The Ohio contest is a Senate special election between appointed Republican incumbent Jon Husted and former Democratic senator Sherrod Brown. Brown's previous statewide victories demonstrate that voters have elected him even as the state has shifted toward Republicans in presidential elections; his 2024 defeat also shows the limits of relying on an older personal vote. Both pieces of history matter, but neither decides the present race.

The difference between preference and expectation

Bowling Green State University's October polling put Brown at 48% and Husted at 42% in the Senate preference question. In the same research, respondents were evenly divided, 48% to 48%, on who they expected to win. This illustrates why perceived electability, personal voting preference and forecast probability are three different things. Neither question is an official result, and both are subject to survey limitations.

Issues and spending

Ohio's campaign has focused partly on industrial and energy policy, household costs, and the growth of electricity-hungry data centers. Campaign and outside-group spending is substantial. Large spending totals indicate the intensity of the contest, but dollars spent cannot simply be converted into votes gained. Analysts should examine whether a candidate's support changes after major advertising and whether the movement is corroborated across surveys.

Putting the race rating in context

A forecast can treat Brown's personal electoral history as meaningful while still accounting for Ohio's more recent Republican statewide baseline and Husted's incumbency. The central question is how much the Senate electorate will resemble the presidential electorate. This is not answered by the presidential margin alone, a single poll, or the candidates' fundraising. The rating above reflects Will's existing site data, not a new prediction supplied by this article.

What could move the rating

Compare later polls from different organizations, ideally distinguishing registered and likely voters. Watch whether undecided voters consolidate and whether the gap between vote preference and expected winner narrows. If polling shifts toward one candidate consistently across methods, that is stronger evidence than one unusually large lead in a survey with a broad margin of error.

The 8.1-point primary result is a significant data point

On August 18, Mary Peltola received 82,244 votes (49.5%) in Alaska's nonpartisan Senate primary; Sen. Dan S. Sullivan received 68,726 (41.4%). That 8.1-percentage-point gap, reported in certified results, is a real statewide election result rather than a poll's estimate. It provides valuable evidence of Peltola's demonstrated support in this particular electorate. However, a top-four primary is not a trial run with identical turnout: the November electorate, ballot and campaign context will differ. Peltola also has a record of competing across party lines, including her 2022 statewide House wins. Those concrete examples help explain why simple comparisons with Alaska's Republican presidential margin can miss important candidate-specific dynamics.

Sullivan's approval: third least popular in January, fifth in July

In January 2026, the Alaska Democratic Party cited Morning Consult data ranking Sullivan the third-least-popular U.S. senator, with 39% job approval and net approval eight points below zero. A July release citing a newer Morning Consult measurement put him fifth least popular with 38% approval, illustrating why the ranking must be dated rather than presented as a permanent fact. Separately, The Hill and Decision Desk HQ's October 14 favorability estimate reported 42.7% favorable and 48.2% unfavorable among Alaskans. Job approval and personal favorability are different questions, and an unfavorable incumbent rating does not by itself translate into a particular Senate vote.

Peltola's statewide appeal goes beyond the Democratic label

Peltola won Alaska's 2022 special and regular U.S. House elections despite the state's long Republican presidential voting record. Her public profile emphasizes fisheries, Alaska Native communities, local decision-making and energy development, rather than requiring voters to agree with every national Democratic position. That history gives the 2026 contest a different candidate-specific context from a generic Democrat-versus-Republican matchup. Calling her the single strongest candidate in the country would require a cross-state measure that is not available; the supported observation is that she has already demonstrated unusual cross-party appeal in Alaska. She also lost her 2024 House reelection bid, so past wins are informative, not a guarantee.

Why an 8.1-point primary margin matters more than an ordinary poll

The August 18 primary was an actual vote: Peltola won 49.5% and Sullivan 41.4%, an 8.1-point difference. A primary result does not have a polling margin of sampling error, so it should be prominent in the analysis. However, it came from a self-selected primary electorate with 16 Senate candidates; the November general-election turnout can be larger and different, and candidates eliminated from the primary may have supporters whose choices change. The responsible interpretation is that Peltola demonstrated substantial real-world support in August, while Sullivan's November support cannot be calculated by simply copying the primary percentages. The primary may be an especially useful signal, but calling it a perfect general-election poll would overstate it.

Independent-minded voters and Alaska-specific issues

Alaska has a particularly large bloc of registered voters classified as undeclared or nonpartisan, according to the state Division of Elections. That registration pattern helps explain why a party label alone can be a poor guide to a candidate's coalition, though it does not mean every unaffiliated voter always votes independently. Fisheries and salmon stocks, fuel and heating costs, the oil industry, rural healthcare, Native community priorities and federal land policy can be especially salient in different parts of the state. Recent Sullivan–Peltola debates addressed energy prices, fisheries, healthcare and foreign policy. Both candidates have their own arguments on those issues; turnout and reliable surveys are needed to determine their actual electoral effects.

What it means that Peltola is the only Democrat in the final four

Alaska's November top-four lineup consists of Peltola, incumbent Dan S. Sullivan, Gerald Heikes and another candidate named Dan J. Sullivan; the latter three are listed as Republicans or Republican-affiliated in election coverage. Democrat David Leslie, who placed fourth in the primary, withdrew, making Peltola the only Democrat remaining. That removes the possibility of a separate Democratic candidate drawing first-choice votes away from her. But ranked-choice voting does not automatically help the lone Democrat: if neither leading candidate has a first-round majority, second and later preferences on eliminated candidates' ballots determine where those ballots go. Supporters of minor Republican candidates may rank the incumbent next, rank Peltola, or leave later preferences blank. The advantage or disadvantage depends on actual ranking patterns, not simply the number of Democrats on the ballot.

Is the race getting bluer as Election Day approaches?

Some 2026 polling has moved toward Peltola as November approaches. For example, a Siena survey reported Sullivan 47% to Peltola 45% in late June, whereas a late-September/early-October Siena survey reported Peltola 50% to Sullivan 43%. Those are different samples at different dates, and other pollsters have measured smaller margins or Sullivan leads. This is evidence of a change in those Siena snapshots, not proof that Alaska always gets more Democratic in the closing weeks. Another phenomenon—Democratic-leaning ballots being counted later during some elections—should not be confused with a change in voter preferences before Election Day. Current measurements and actual turnout remain more informative than an assumed automatic late shift.

How to interpret Will's published rating

The live site displays Will's own Alaska prediction and polling averages. The documented ingredients behind a thoughtful interpretation are Peltola's decisive primary first-place finish, her previous statewide crossover appeal, Sullivan's unfavorable approval and favorability readings, Alaska's independent registration pattern, and the uncertainty of ranked-choice transfers. The countervailing evidence is that Alaska continues to vote Republican at the presidential level, Sullivan has been elected statewide before, Peltola lost her 2024 House race, and the general electorate is not the primary electorate. The most useful next evidence is a transparent likely-voter poll with separate first-choice and ranked-choice-transfer questions, together with independent corroboration of any late trend.

The baseline versus an open seat

Iowa has voted Republican in recent presidential elections, but this year's Senate contest does not feature the incumbent senator seeking another term. Open seats remove part of the personal incumbency advantage and can create an electorate that evaluates two candidates without the same history of voting for either in that office. It is still important not to overstate what that change means: the state's existing party preferences remain a separate piece of evidence.

A clear example of sample differences

Marist's October 8–11 poll found Democrat Josh Turek at 48% and Republican Ashley Hinson at 45% among likely voters, with a reported margin of sampling error of plus or minus 6.6 percentage points. The same poll found a larger Turek margin among registered voters. Those are different populations. A registered-voter sample includes people who may not vote, while a likely-voter screen tries to anticipate actual participation. Neither approach is infallible.

What else the Marist survey reveals

Marist also reported that many respondents expected Hinson to win despite reporting a close race in their personal preferences. That contrast can reflect perceptions of Iowa's underlying partisanship, not necessarily an actual future vote. The campaign includes debates over tariffs and economic pressures. To understand those issues empirically, separate what the candidates emphasize from what representative voters say matters most.

The forecasting challenge

A close Senate survey in a state with a Republican presidential baseline creates two competing signals. A thoughtful explanation of Will's rating should say whether the live model weights the latest candidate-specific data heavily or deliberately leaves room for turnout and historical error. One poll near the midpoint of its margin of error is not enough to prove a lasting change in the state's electorate.

What could change the forecast

Look for new Iowa likely-voter polls after the Marist field dates, especially if they disclose sample construction and how undecided voters are treated. Compare independent pollsters instead of treating every poll as equally informative. A persistent multi-poll trend, or meaningful evidence about actual voter participation, is more useful than a single survey swing.

Why the candidates bring different statewide histories

North Carolina has been closely contested in national elections while also electing statewide candidates from both major parties to some offices. Former Democratic governor Roy Cooper has a record of winning statewide; Republican Michael Whatley has a different political background. A Senate contest is not a replay of a governor's race, but Cooper's previous campaigns provide evidence worth considering when interpreting his current polling.

How to interpret the larger polling margins

An October SurveyUSA survey published a 54% to 40% Cooper–Whatley result among likely voters. Other polls have produced different margins. A double-digit lead in a survey is a fact about the respondents and the poll's assumptions, not a certified vote margin or a precise probability. Poll aggregation can help reveal whether the result is an outlier, but the average depends on which polls qualify and how they are weighted.

Turnout matters even with a substantial polling lead

North Carolina's electorate contains rapidly growing communities, long-standing regional differences and voters with varying party loyalty. Early-voting activity may reveal participation, but it does not automatically show which candidate is benefiting. Vote-method differences, age composition and likely-voter assumptions can matter more to the final estimate than a single national political headline.

The forecast-versus-poll question

If Will's site gives the race a rating that appears more cautious than the latest poll margin, the key distinction is between a measured lead and uncertainty around a future electorate. Incumbency is not the central factor here because the seat is open; candidate familiarity, turnout and the consistency of separate polls may matter more. The website should explain those factors without implying it has measured effects that the data do not support.

What to watch next

Check whether Cooper's measured advantage is consistent across multiple likely-voter pollsters and whether undecided voters and sample composition shift. Be careful with comparisons between a poll of all registered voters and one of likely voters. The strongest case for changing a forecast is a pattern supported by updated, comparable evidence, not one surprising number.

How Georgia became a closely watched Senate state

Georgia's recent statewide races have featured very tight margins and substantial variation in turnout across metro Atlanta, its suburbs and the rest of the state. Jon Ossoff is the sitting Democratic senator, while Republican Mike Collins is a member of the U.S. House. Incumbency and name recognition matter as context, but neither should be treated as a substitute for evidence about this particular election.

What recent polling can and cannot say

Recent public polls have reported an Ossoff lead, although individual polls have different field dates, samples and margins. A poll average is useful for summarizing a set of measurements, but even a consistent lead cannot guarantee the distribution of votes on Election Day. Analysts should distinguish whether a poll asks registered or likely voters and whether it includes all options voters will see on the ballot.

The campaigns' substantive disagreements

In their October debates, Ossoff and Collins presented different accounts of tariffs and their effects on affordability and manufacturing. They also disagreed over federal healthcare policy and the war in Iran. Reporting on those positions helps explain the policy stakes, but whether they change voting behavior is an empirical question. A forecast should focus on how those disputes relate to actual surveys, past turnout and undecided voters.

Why statewide margins deserve caution

Georgia's electorate is not uniform. Small shifts in a few large communities, or turnout differences between presidential and midterm years, can change a close statewide result without a broad change in every county. Will's rating above is his published call. A defensible explanation should identify the limits of transferring a past presidential result into a Senate forecast, and avoid interpreting fundraising as votes.

What could change the assessment

Watch polling after the debates, including how large the undecided share remains. Compare surveys from different organizations and their sampling populations. Actual early-voting totals may provide context about turnout, but they do not reveal voters' choices by themselves. If evidence moves in a consistent direction across sources, the rationale for updating the published rating becomes stronger.

What an open seat changes

The retirement of Democratic senator Gary Peters makes Michigan's 2026 Senate race an open-seat election between Democrat Abdul El-Sayed and Republican Mike Rogers. With no Senate incumbent defending the seat, the candidates' own histories, campaign organizations and messages may matter more than an incumbent's established constituent-service record. But the state's close recent presidential results remain important context for judging unusually large or small survey margins.

The polling signal and its limitations

Recent Michigan surveys have put El-Sayed and Rogers only a few points apart. For example, an October Rasmussen survey reported El-Sayed at 46% and Rogers at 44%; different pollsters have found different margins. A narrow lead is useful evidence about the sample and field period, not a prediction that it will persist. Comparing candidate preferences with the voting population modeled by each poll is essential.

Why the issues are locally distinctive

The campaigns have addressed economic conditions, employment in Michigan's automotive industry, foreign policy and their political records. Rogers previously served in Congress, while El-Sayed has a public-health and policy background. Those differences help explain the contest's lines of argument, but no analyst can reliably convert a candidate biography or a debate sound bite directly into a percentage of the final vote.

A useful caution from Michigan's primary

Michigan's Democratic primary illustrated how polling can vary significantly from a final close result. That does not prove that general-election polling will miss in the same direction or by the same amount: primary and general electorates differ. It does show why Will's explanation should separate polling averages, candidate history and assumptions about which groups will actually turn out.

What would justify a model update

Track multiple late likely-voter surveys, their field dates, sample types and undecided shares. Treat reported advertising expenditures as campaign context, not an automatic change in vote margin. A clear, sustained movement across independent polls is more persuasive than any single result. The most transparent forecast records when its rating changed and which new evidence prompted the change.