Senate Race Report: September 20, 2026

Question: Seven weeks out, with Kansas handing Adam Hamilton his first independent lead of the cycle, Kalshi’s own desk reporting the Democrats at a record 57 percent for the Senate, and the pollsters calling Minnesota, Michigan, and New Hampshire all closer than the models do, is the majority closer than it was a week ago?

Answer: The models and the markets moved decisively Democratic this week, and this time the forecasters moved a little with them. Kalshi’s Senate-control market crossed to the Democrats on September 14 and Kalshi’s own desk put them at a record 57 percent by September 17, their first extended Democratic price since April, with Polymarket sitting near the same number.¹ Nate Silver’s FLIPR model rose from 50.6 percent on September 6 to 58.6 on September 14 and into the low sixties by the weekend, and Decision Desk HQ’s simulation ended the week with a median 51-49 Democratic Senate.²³ The forecasters who do not model moved only twice, and both times toward a toss-up: Cook made New Hampshire one on September 15, and Inside Elections made Maine one on September 17.⁴ The ground moved with the models: the first independent survey of the cycle put Hamilton ahead in Kansas,⁵ Minnesota went dead even,⁶ El-Sayed opened his largest Michigan lead of the cycle,⁷ and Florida’s special election tied in a poll the Moody campaign answered with an emergency fundraising alert.⁸

Democrats still need a net gain of four seats, and a 50-50 Senate belongs to Vance. The path still runs through North Carolina, two coin flips in Ohio and Texas, and a bench from Maine to Alaska. The margin seat led a poll this week for the first time.

What Changed This Week

The week’s frame was set by the markets, and this time they did not fade. Kalshi’s Senate-control market crossed to the Democrats on September 14 for the first time since April, touched 53 percent the next day, and stood at 57 percent by September 17, with Polymarket close behind; Kalshi’s own news desk called it an all-time high for the Democrats.¹ The same environment that moved the market last week kept pushing: Brent crude traded above $107 a barrel on September 14 with West Texas Intermediate above $103, Goldman Sachs warned that a continued fight could take oil past $120, and traffic through the Strait of Hormuz has collapsed to single digits of ships a day from more than a hundred before the war, with war-risk insurance now running to 10 percent of a hull’s value per transit.⁹

The models moved with the markets, and Silver’s went furthest. FLIPR rose from 50.6 percent for a Democratic Senate on September 6 to 58.6 on September 14, and its Deluxe build sat near 65 percent by September 19; the model also moved Michigan from lean to likely Democratic and Kansas from likely to lean Republican during the week.² The Silver Bulletin generic ballot, at Democrats plus 7.5 with a likely-voter adjustment of plus 8.2, is the highest of the cycle.¹⁰ Decision Desk HQ’s simulation gave the Democrats 52 percent control odds on September 18 and a median simulated majority of one seat.³

Kansas, the margin seat, had the week’s biggest single development. Emerson College’s survey, in the field September 15-16 and released September 18, shows Hamilton at 45 percent to Marshall’s 43, with 8 percent undecided; it is the first independent survey of the cycle to put the Democrat ahead, and it went into the field two days after Vice President Vance campaigned for Marshall in Olathe.⁵ The markets still price Marshall around 71 to 74 percent, and the ratings tables still call the race “likely” Republican, but Silver’s model now calls it “lean,” and Charlie Cook’s column on September 17 listed Kansas among the nine genuinely competitive races in the country, calling it the cycle’s biggest surprise.⁴ The honest caution remains the one this report has repeated: Kansas produced competitive September polls in 2014 and 2020, and the Republicans won both Novembers by double digits.¹¹

The defense tightened in the polls. Minnesota moved the furthest: SurveyUSA’s survey released September 15 found Flanagan and Michele Tafoya dead even at 42, with 12 percent undecided, and a co/efficient survey taken September 14-15 gave her a one-point edge, a month after the state was a double-digit walk.⁶ Michigan’s two new surveys, the Washington Post-Schar School poll released September 17 and Emerson’s from the same week, both put El-Sayed up three and two points, and Rogers pulled out of a Detroit debate this week, leaving El-Sayed to address an empty podium.⁷ New Hampshire’s new co/efficient survey, in the field September 9-11, tied the race at 46, and Cook’s decision to move the state to a toss-up on September 15 ratified the polling.¹² Georgia went the other direction: Quantus’ survey released September 18 has Ossoff up 48.2 to 44.4 after a tie a week earlier, Rasmussen has him up nine, and the Atlanta Journal-Constitution reported the race slipping off the GOP’s radar as promised Republican television arrives slowly.¹³

Florida produced the long tail’s biggest surprise. St. Pete Polls’ survey, in the field September 15-17 and released September 17, tied the special election at 45 apiece, the first public reading without a Moody lead, and Moody’s campaign answered with an emergency fundraising appeal calling Nixon a threat; the appointed incumbent has so far declined debate invitations.⁸

Ohio and Texas stayed coin flips, all leaning left of even in the new surveys. Bowling Green State’s YouGov survey, released midweek, has Brown up 48 to 45 while 53 percent of Ohio voters say they expect Husted to win; no Senate debate is scheduled, and Husted’s campaign has said it would be shocked if Brown debated, given his 2024 refusal.¹⁴ Texas produced a week of fresh numbers all on Talarico’s side: Emerson has it 47-46, Mason-Dixon 46-43, and Texas Pulse 49-43, while his late-night television appearance drew twelve million views.¹⁵

North Carolina’s polls slid further from the Republicans. The Carolina Journal’s Harper Polling survey, in the field September 13-15 and released September 17, shows Cooper at 49 to Whatley’s 34, the widest margin in that series, with the Republican holding just 74 percent of his own party and undecideds doubled.¹⁶

Alaska, Iowa, and Nebraska produced contradictory reads, as usual. AARP’s survey released September 17 has Peltola up 46 to 41, and 53-47 after ranked-choice reallocation; co/efficient’s September 14-17 survey has Sullivan up 48 to 46.¹⁷ Iowa’s new co/efficient reading gives Hinson 46 to Turek’s 42, a week after Rasmussen put Turek up five and days after Cygnal tied them at 43.¹⁸ Nebraska’s two new surveys disagree by eight points of margin: SurveyUSA has Osborn up 46 to 42, Wedgewood has Ricketts up 52 to 48.¹⁹

The tail had a busy week. South Carolina’s Rasmussen survey, released midweek, has Graham up 48 to 43, and the candidates finally agreed to a single October 11 debate on state public television.²⁰ Montana’s Quantus survey, released September 17, has Alme at 43.6 percent with Bodnar and Bankhead splitting the Democratic field, and the models give Alme about nine in ten.²¹ In Louisiana, Davis worked an LSU voter event while the two Republicans Letlow beat still will not endorse her.²² Mississippi’s television is the story there: about $4.6 million in combined spending, under a fifth of the 2020 race’s pace, and Hyde-Smith still will not commit to debating Colom.²³

The environment kept moving the Democrats’ way. Reuters/Ipsos gives them a seven-point generic ballot lead, Cygnal’s early-September reading nine, and the Silver Bulletin average sits at plus 7.5, up from 6.4 at the start of the month; the same Ipsos survey puts the president at 35 percent approval.¹⁰

The Defense

Michigan is still the most competitive Democratic seat in the country, and it is the one where the week gave the party genuine news. The Washington Post and Emerson surveys both put El-Sayed ahead by two to three points, his best readings of the cycle, though the Post also found 53 percent of likely voters want a Democratic Senate, which says the nominee is underperforming his party’s ceiling; Rogers skipping the Detroit debate let El-Sayed own the stage alone, and the empty podium is the image the campaign will keep.⁷

Georgia is the second seat, and it quietly got safer. Quantus went from a 38-38 tie last week to a four-point Ossoff lead, Rasmussen has nine, and the Journal-Constitution’s finding that the GOP’s promised television has been slow to arrive suggests the party’s own models are not treating the state as live, which is how a race becomes safe without any forecaster saying so.¹³

Minnesota is the new worry. Two September surveys have it even, a dead heat in a state the models price at nine in ten for the Democrats; the polls have moved more in Minnesota in a week than the ratings have all cycle, and a defense seat that is actually a coin flip changes the arithmetic on every path.⁶

New Hampshire is the fourth, and Cook’s move made the forecasters honest about what the polling says. The co/efficient tie at 46 and Cook’s September 15 shift to a toss-up describe the same race: a former two-term senator running even in a state where the other party’s governor leads the ticket.¹² Decision Desk HQ still prices Pappas at 78 percent, but the two readings of the state that agree with each other both have it even.

The other nine Democratic seats, Colorado, Delaware, Illinois, Massachusetts, New Jersey, New Mexico, Oregon, Rhode Island, and Virginia, are not in play.²⁴

The Offense, Tiered

North Carolina remains the anchor of every path, and the week made the case easier, not harder. The Carolina Journal series, run by a Republican-leaning firm, delivered Cooper’s widest lead of the cycle at fifteen points, with Whatley’s support eroding into undecided; the markets have held Cooper above nine in ten all month.¹⁶

Ohio and Texas are the two coin flips, and both lean left of even in the newest polls. Brown is up three in the BGSU/YouGov survey in a race where cost of living and the war are the top issues, but 53 percent of Ohio voters expect Husted to win, and the outside money is near parity, about $80 million spent for the Republican against $69 million for the Democrat; the firewall is structural, not rhetorical.¹⁴ Texas is the week’s liveliest poll page: Talarico leads the week’s surveys by zero to six points, the Republican super PACs have reserved the largest battleground buy of the cycle to defend Paxton, and a state that has not elected a Democrat to the Senate since 1988 is polling like the tipping point it is, with the voter registration deadline October 5 and early voting October 19.¹⁵

Maine, Alaska, Iowa, and Nebraska are the bench, and three of the four now poll like races. Maine is the race a forecaster moved this week, Inside Elections’ September 17 shift to a toss-up joining Cook and Sabato, and the agreed four October debates all come after early voting opens October 5, a schedule Collins fought; six write-in candidates could even trigger ranked-choice counting in a state she survived blue waves in before, in 2008 and 2020.²⁵ Alaska’s two new surveys point different ways, which is exactly what ranked-choice voting and a same-named spoiler do to a race; the Democrat won the primary by eight points, and the AARP survey’s reallocation gives her the general by six.¹⁷ Iowa keeps alternating by ten points between pollsters, and the models have not budged; it is a coin flip on Republican fundamentals.¹⁸ Nebraska is the wildcard: Osborn leads in one new survey and trails in the other, in a state the president carried by twenty points, and an independent win there is the cheapest possible fourth seat because it needs no party apparatus at all.¹⁹

Kansas is the margin seat, and this week it led a poll. Emerson’s survey, fielded two days after the vice president held the Republican’s rally in Olathe, is the first of the cycle to put Hamilton ahead, and Charlie Cook’s list of nine competitive races finally includes the state, which he called the biggest surprise in the country.⁴ The markets still price Marshall at 71 to 74 percent, a reminder of how far the state leans; but a race where the party in power sends the vice president and a Democratic pastor draws even in the first independent survey is a margin seat in fact, not just in theory.⁵

Florida stopped being a foregone special election this week. St. Pete Polls tied it at 45, Moody’s campaign is sending emergency fundraising appeals in a race she outspends Nixon by more than ten to one,²⁶ and the appointed incumbent has yet to accept a debate invitation; the forecasters and markets still lean Republican, but the warning shot is in the record.⁸

Louisiana, Mississippi, Montana, and South Carolina remain Republican-favored. Letlow leads the Louisiana money race about three to one and the models by ninety, while the two men she beat refuse to endorse her and Davis works the campuses; Hyde-Smith leads Mississippi by six and a half in the average while her spending runs at a fifth of the pace of her last race; Alme leads Montana’s four-way field at 43.6 with a ninety-one percent model; and Graham leads South Carolina’s new poll by five with the agreed October 11 debate on the calendar.^22232120

The nine safe Republican seats, Alabama, Arkansas, Idaho, Kentucky, Oklahoma, South Dakota, Tennessee, West Virginia, and Wyoming, are not in play.²⁴

The Math

The forecasters and the markets now disagree in a new way: the tables are nearly still, and the models and markets are not. The consensus ratings moved twice this week, Cook’s New Hampshire shift on September 15 and Inside Elections’ Maine shift on September 17, both toward toss-up; no race changed hands in a ratings column. The models and markets moved much more: Silver’s FLIPR sits in the low sixties for a Democratic Senate, Decision Desk HQ’s simulation has a median 51-49 Democratic Senate at 52 percent odds, and Kalshi’s own desk put the Democrats at a record 57 percent on September 17, with Polymarket near it.²⁷ The split is the story: the forecasters read the same polls the models do and are choosing not to move whole races on them, while the markets are bidding on the same polls and moving every day. Two straight weeks of model-and-market movement without corresponding ratings moves is not a quiet market; it is the models pricing what the tables are still refusing to say.

The market path is North Carolina, Ohio, and Texas, plus one of Maine, Alaska, Iowa, or Nebraska. That is four flips, and it is what the markets are pricing.

The consensus path is North Carolina, Ohio, Maine, and Alaska or Iowa. It requires winning the top four.

The nightmare is any loss on the defense, and the defense polls made the nightmare less hypothetical this week: Minnesota dead even, Michigan up two to three, New Hampshire a Cook toss-up. Every one of those losses would add a fifth flip, and the model set would need to be wrong in the same direction everywhere at once.²⁸

The Kansas connection is worth stating plainly again. If the Democrats sweep the top three, Kansas is not needed. If any of the three slips, Kansas, or Nebraska through Osborn, becomes the margin seat. This week Kansas led a survey for the first time, the forecasters’ own dean listed it among the nine competitive races in America, and the party in power sent the vice president to a state it had written off for a generation. That is the margin seat becoming real.⁴

The ratings table

Ratings as of September 20, 2026, with each column on its own schedule: Cook Sep 15; IE Sep 17; Sabato Aug 26; RCP Sep 18; DDHQ Sep 18; Silver Sep 19.

A word on what the columns mean. PVI is the Cook Partisan Voting Index, and it is the only backward-looking number on the table. It measures how far a state leans from the nation as a whole, computed from the 2020 and 2024 presidential two-party vote shares weighted toward 2024; it is denominated in vote share for one party rather than margin, so it runs about half the size of a margin-style lean (Cook calls Texas R+6; Silver’s partisan lean calls it R+11). Cook says it himself: race ratings reflect an outlook for which party will win the next election, while the PVI takes a longer view. So it earns its place as contrast, not as forecast: “Tossup in an R+6 state” only means something against that background. Across the whole map, lean is a strong prior, Pew found 158 of 176 Senate races since 2012, 89.8 percent, went the way of the presidential carry; but applied to exactly these nine races it calls only five of nine correctly, missing Georgia, North Carolina, Ohio and Texas, all of which it would call Republican and all of which the forecasters now favor for the Democrat. That is not a contradiction, it is the selection effect: a race earns its way onto this table precisely by being a case where the baseline does not settle it. The counter-dimension is elasticity, how hard a state swings when the nation swings, and it is nearly orthogonal to lean (r = +0.12 across these nine). Maine and New Hampshire are among the most elastic states in the country, which is part of why they are toss-ups; Georgia and North Carolina are among the most inelastic, which is part of why forecasters rate them Democratic despite Republican-leaning PVI numbers. IE is Inside Elections, Nathan Gonzales’s nonpartisan newsletter, one forecaster among several. The rating scale runs solid, likely, lean, toss-up, with IE’s Tilt slotted between lean and toss-up; Cook, Sabato, and DDHQ do not use Tilt, so a Tilt appearing only in the IE column is not disagreement, it is a finer scale. (flip) marks a rating that forecasts a change of party from the current holder.

A note on the sources. One column is arithmetic and six are not. PVI alone can be reproduced from certified FEC totals, and we have reproduced it, matching all nine of Cook’s published figures; Cook, Inside Elections, Sabato and RCP are analyst judgment, which means their ratings cannot be independently reproduced from any underlying data, and RCP’s method in particular is unverified to us because the site blocks automated reading, so treat that column as unlisted rather than as a seventh opinion. DDHQ and Silver are published models, but the columns are not independent: Silver’s FLIPR feeds Cook, Inside Elections and Sabato ratings in as an input at roughly one-sixth weight, so where his column agrees with theirs the agreement is partly built in, and all six read the same polls and talk to the same operatives, which means six agreeing columns are not six confirmations. The signal is where they disagree, currently Georgia, Iowa, Ohio and Texas, and freshness differs per column and is stated above, because Sabato’s ratings are three weeks staler than Cook’s.

The Close

The report lands seven weeks from Election Day with the markets doing what the forecasters would not: making the majority a Democratic bet for the first sustained stretch of the cycle. The war that set the frame last week kept pushing, oil above $107, Hormuz traffic in single digits, freight costs at records; the generic ballot reached its cycle high for the Democrats; Kansas gave the margin seat its first lead; and the defense polls moved closer everywhere except Georgia.

The honest cautions still hold. The ratings barely moved, because the movement is running through the models, the markets, the polls, and the money. Minnesota, Michigan, and New Hampshire all poll closer than the models say. Kansas has been here in September before, in 2014 and 2020, and the Republicans won both by double digits. A model at 65 percent and a market at 57 percent are not a win; they are a price, and prices move both ways for seven more weeks.

But the direction is no longer arguable. When a state like Kansas leads a Democrat in an independent survey in September, when the party in power sends the vice president to hold it, when a special election in Florida ties in a poll, the map is not the map of the spring. The math is narrow and unchanged: hold the defense, flip North Carolina, win two of the coin flips, and find a fourth seat somewhere in the long tail. This week the fourth seat led a poll in Kansas, tied in Florida, and went even in Minnesota. That is what movement looks like sixty-one days out.

PRH | huffmanwrites.org | © Philip Huffman

Notes