Analysis

Marist Texas polls shift both statewide forecasts toward Democrats

Marist interviewed 1,139 registered Texas voters September 17–20 (±4.4 points). James Talarico leads Ken Paxton 50%–44% in the Senate test, with 2% choosing another candidate and 4% undecided or refusing. Gina Hinojosa leads Greg Abbott 49%–46% in the governor test, with 1% choosing another candidate and 3% undecided or refusing. Both polls are enabled in the workbooks and included in this release. Older polls were reweighted using the September 22 model cutoff.

Polls added

Simulation reruns

The 50,000-draw Senate rerun projects 51.67 Democratic caucus seats on average (median 52; middle 80%: 49–55), with a 70.96% Democratic chance of control and 29.04% Republican chance. The 10,000-draw seeded House rerun projects 234.74 Democratic seats on average (median 233; middle 80%: 205–266), with a 75.05% Democratic chance of control. All 435 House district estimates and Senate and governor county maps use model inputs through September 22. County and district baselines and geometry are retained. No governor control probability is published.

The state-by-state change log compares the prior public release with this September 22 model refresh. Texas changes include the new Marist poll; other movements reflect the refreshed forecast inputs and poll weights.

New polling nudges ten Senate, governor, and House estimates

This release adds 13 previously missing polls: seven Senate, five governor, and one House poll. The 12 statewide polls bring the Senate and governor archive to 437 records; the House archive now contains 115 polls. All 13 are enabled in their respective workbook averages. The table below compares the September 19 statewide estimates with the updated workbook outputs; VA-01 is compared with its pre-poll district estimate.

Polls added

Race margins use Democratic minus Republican percentage points; positive values favor Democrats. The Senate control simulation remains the frozen September 19 snapshot: Democrats 48.3%, Republicans 51.7%, with 50.3 expected Democratic caucus seats from 500,000 draws. No governor control probability is published. The House control simulation remains the provisional September 6 snapshot (10,000 runs): Democrats 74.9%, Republicans 25.1%; the House point estimate update does not rerun it. County and district maps use the refreshed statewide or district estimate while retaining their existing geographic baselines.

Forty workbook poll rows added as the national environment moves to D+8.65

The September 17 workbook update added 40 poll rows. Two duplicate Alaska rows were removed during validation, leaving 38 unique new polls in the published snapshot. The national generic congressional ballot input is now D+8.65, which shifts the pre-poll environment toward Democrats before state polling is blended in.

The race changes combine the new polls with the GCB shift and are read from the workbook’s live snapshot. The simulation is marked LIVE VOLATILE in the workbook, so these probabilities and seat percentiles are recorded context rather than a frozen forecast.

Five recommended polls stay enabled; two archive entries remain excluded

The five enabled recommended polls and their recorded workbook effects are listed below. Each transition is polling control OFF → ON, holding the other added polls ON. The Montana Peak Insights/Alme poll and Florida Change Research poll remain visible in the archive and state pages, but both controls are off in the live workbook.

The race-margin changes above are deterministic workbook effects. The live simulation is a refreshed volatile snapshot, not a frozen forecast. Poll source links: NH · KS · IA · KY · SD · MT · FL.

Five new polls leave control of the Senate a true toss-up

We added five new general-election polls from the current RealClearPolling Senate list. CNN has Troy Jackson ahead of Susan Collins 48% to 45% in Maine and Abdul El-Sayed ahead of Mike Rogers 47% to 44% in Michigan. Detroit News/Glengariff has Rogers ahead of El-Sayed 46% to 44%. Univision/YouGov has James Talarico ahead of Ken Paxton 48% to 43% among registered voters in Texas. TIPP has Peggy Flanagan ahead of Michele Tafoya 44% to 42% in Minnesota.

The two Michigan polls were enabled together, so the workbook gives their combined state effect rather than pretending that a volatile live simulation can provide a clean isolated number for each one. Together they moved Michigan’s final margin from D+3.30 to D+2.92, a 0.38-point Republican shift. The Maine poll leaves the race at Tilt D. The TIPP poll moved Minnesota’s final margin from D+7.91 to D+6.52, a 1.39-point Republican shift. Texas moved from D+0.13 to D+0.41, a 0.28-point Democratic shift. The live simulation moved to 48.9% Democratic control and 51.1% Republican control. RANDARRAY results are volatile; this is the recorded September 9 snapshot, not a permanent probability. RCP Senate poll list →.

Candidate history now survives the final poll blend

The old model had two related problems. It could throw away the strongest part of a well-established candidate record when the polls were blended in, and it gave every major independent the same three-point allowance. Version 0.52 deals with both. A candidate who hits the exceptional history cap can now keep another 0.93 points after the poll blend. A candidate with at least three consistent statewide results can keep one point, provided the underlying candidate score is at least three points. Either way, the extra adjustment is capped at one point.

For independents, we now use the candidate’s own statewide record first. Dan Osborn gets half of his weighted 2024 overperformance, minus the candidate-strength credit already in the model. Brian Bengs has no comparable independent run, but he does have current Senate polling, so the three-point provisional allowance still applies. Todd Achilles has neither. He no longer gets the allowance. Regular third-party candidates do not qualify. These rules directly changed five races and left the other 30 alone. In one 50,000-run check, Democrats won Senate control 48.5% of the time and Republicans 51.5%. That is a live RANDARRAY result, not a frozen forecast; refresh the workbook and it will move.

New polling moves Iowa, Tennessee, and New Mexico

We added three polls. Emerson has Ashley Hinson leading Josh Turek 50% to 45% among 750 likely voters in Iowa. Beacon Center has Bill Hagerty ahead of Marquita Bradshaw 58% to 33% among 1,200 likely voters in Tennessee. And the Albuquerque Journal poll has Ben Ray Luján leading Larry Marker 53% to 38% among 516 likely voters in New Mexico. The Suffolk Iowa poll was already in the workbook, so we did not add it twice. The Texas change was a correction, not a new survey: Overton/TPPF’s cumulative result, including leaners, was 50–50.

None of the four races changed categories: Iowa is still Tilt R, Tennessee Safe R, New Mexico Safe D and Texas a Toss-up. The race-level changes are more useful than comparing two live simulation draws, since RANDARRAY moves every time the workbook recalculates. Emerson · Beacon · New Mexico Journal poll.

Silver Bulletin D+6.6 becomes the national baseline

We switched the Senate model’s national environment from our own D+5.8 estimate to Silver Bulletin’s D+6.6 generic-ballot average. The model applies 75% of that difference, so the change adds 0.60 points to each race before state polls enter the calculation. Races with a lot of polling move less because the polls carry more of the final estimate.

Georgia moved from D+6.59 to D+6.87, while Maine moved from Tilt D to Lean D. In the frozen 500,000-run simulation made with this update, Democrats won control 46.1% of the time, up 3.1 points. Republicans were at 53.9%. The model averaged 50.19 Democratic caucus seats, with the middle 80% of outcomes still running from 47 to 53.

A softer national environment trims Democratic margins

Before the switch to Silver Bulletin, our generic-ballot input fell from D+6.4 to D+5.8. Since the model uses three-quarters of the national margin, that cut 0.45 points from every race before state polling. The most heavily polled races barely moved; their polling averages had not changed.

Georgia moved from D+6.86 to D+6.66. None of the races shown above changed ratings from this adjustment alone.

Two new polls leave both races on a knife edge

Two polls were added after the generic-ballot update. Alaska Survey Research has Mary Peltola ahead of Dan Sullivan 50.6% to 49.4% among 1,495 likely voters. Overton Insights has James Talarico leading Ken Paxton 50% to 45% among 1,167 likely voters.

In the frozen 500,000-run simulation made at the time, Republicans won control 56.9% of the time and Democrats 43.1%. The average chamber was almost exactly tied: 50.01 Democratic caucus seats and 49.99 Republican seats. The middle 80% ran from 47 to 53 Democratic seats.

New Michigan poll modestly improves the Democratic position

MSU’s Office for Survey Research and IPPSR has Abdul El-Sayed ahead of Mike Rogers 50% to 45% among 779 likely voters. The survey used a YouGov panel. We list MSU/IPPSR as the pollster and give it the model’s neutral 1.00 multiplier because the organization is not separately rated.

Michigan stayed Tilt D. In the accompanying 500,000-run snapshot, Democratic control rose from 48.2% to 48.3% and the average Democratic caucus moved from 50.292 seats to 50.299. In other words, the poll helped, but hardly by enough to rearrange the Senate. The middle 80% ran from 48 to 53 Democratic seats. Read the release →

Wedgewood nudges Iowa toward Republicans

Wedgewood Polls has Ashley Hinson ahead of Josh Turek 50% to 48% among 600 likely voters. Wedgewood is not rated, so the model gives the poll a neutral multiplier and its full 1.000 base weight.

EPIC-MRA moves Michigan toward Democrats

EPIC-MRA has Abdul El-Sayed ahead of Mike Rogers 50% to 45% among 600 likely voters. The result moved Michigan’s polling average almost a full point toward Democrats, though the final forecast moved less because the model still retained its fundamentals estimate.

At the time, the displayed margin moved from D+2.4 to D+3.1. Later polling pushed it to D+3.37. Michigan remained Tilt D.