Gas Prices and the 2026 Midterm Battlegrounds

How to read the headline. The Senate chance of control comes from polling alone, after the 3.5-point correction toward Republicans described below; the House shows only a rough seat count, because 407 of its 435 seats are not modelled here. We simulate the election 100,000 times. In each run every race misses its polling average by a random amount, made of a national miss shared by all races (typical size ±3 points) and a race-specific miss (±3.5 points for Senate races, ±5 for House districts, which have no district polling). The headline reports how often each party wins a majority, the most likely seat count, and the range that covers 80% of runs. The win chance column is how often each race’s leader wins across the same runs — a D+2 race is closer to a coin flip than the margin suggests.

Ratings. A race within 3 points is a toss-up, whoever is narrowly ahead — that is inside normal polling error. 3–6 points is Lean, 6–10 Likely, 10+ Safe.

The gas-price scenario is a “what if”, not a forecast. Pump prices are up far more than in a normal midterm year: both figures are measured from the EIA’s weekly national gasoline price, comparing this year’s rise since early January with the average rise to the same date in the eight midterm years 1994–2022. The scenario turns that excess into one uniform swing toward the party out of power (0.04 points per excess point, capped at 4) and applies it to every race. That conversion is an assumption: research on fuel prices and vote choice is weak and inconsistent, and voters in different states will not react identically. Today’s polls also already reflect today’s prices, so the scenario partly double-counts. Read it as “what if the pump keeps biting from here”. It is off by default.

The Combined scenario (added 8 October, turnout and affordability revised 10 October, nominees added 13 October). Six ingredients, with overlaps removed so the same unhappy voter is not counted twice. Gas: half the pump swing above plus half a state-by-state model of how voters react to the pump price over the last 60 days (EIA national weekly price). Road-trip interviews with seven 2024 Trump voters (7 Oct) showed most staying loyal and the unhappy ones more likely to stay home than switch, so the model has the GOP base lose 0.25 points of support per 10¢ rise, 80% of it by staying home and 20% by voting Democratic (which counts double), and independents move 0.4 points per 10¢. It is scaled by the share of workers who drive alone to work (ACS 2024: states for the Senate, districts for the House, against 69.2% nationally), and in the seven 2024 swing states (AZ, GA, MI, NV, NC, PA, WI) party turnout operations are assumed to win back 30% of the stay-at-homes. Falling prices work the same way in reverse. This replaces the 8 October version, which assumed every Trump voter behaved like the seven interviewees. Approval drag, +0.6: Harry Enten (CNN, 7 Oct) has Trump at −21 net in the key Senate states against Biden’s −13 in 2022. That is scored at 0.15 points per approval point, at half weight because part of it is cost of living already in the gas line. GOP turnout, about +1.7 (+1.2 in the seven swing states), revised 10 October: 76% of Harris 2024 voters say they are almost certain to vote against 63% of Trump 2024 voters (Reuters/Ipsos, Oct). Assuming 87% of the certain and 40% of the rest actually vote, that is about 76% turnout against 70%, a 6-point gap. In 2018 and 2022 the out-party’s voters turned out 4 points more than the president’s (Pew validated voters), so this is a worse-than-normal midterm for the GOP; Pew’s October motivation gap (63% v 46%, reversed from 2022) and Reuters/Ipsos enthusiasm (48% v 30%) point the same way. The gap is not in the MAGA core: Marquette (2–9 Sep) has Republicans very favourable to MAGA 93% certain or very likely to vote, against 74% of Republicans unfavourable to it. Because the shortfall sits with soft Republicans, the voters turnout operations chase, the seven 2024 swing states get the same 30% win-back as the gas model. 60% of the gap is assumed already in the polls, since most October polls screen for likely voters. Affordability, about −0.1: conservative outlets now mention “affordability” almost as often as liberal ones (56 v 60, against 41 v 82 in August; WSJ count), so the Democrats no longer have the cost-of-living message to themselves. The gas line is trimmed by 11%: half the fall in the Democratic share of that talk, the other half being what voters see at the pump. Unpopular GOP nominees, 13 October: Harry Enten’s aggregate (CNN) has five Republican Senate nominees underwater: Mike Rogers (MI) −8, Ashley Hinson (IA) −9, Dan Sullivan (AK) −11, Jon Husted (OH) −13 and Ken Paxton (TX) −16. The head-to-head polls already reflect that, so the only addition is how each race’s undecided voters (5–9% in our weighted averages, updated daily) break: they are assumed to split the way the whole electorate rates the nominee, so Paxton’s −16 sends Texas’s undecided 58–42 to Talarico. That is worth about +0.5 to +1.1 points. No ratings are given for the Democratic nominees, who are treated as even; if they are underwater too, this overstates. Registration, pro-GOP: Enten’s GOP registration gains since 2018 (NC +8, PA +7, AZ +4) at 0.1 points each. Not applied, because they are already inside the approval figure and the generic ballot: Trump at −42 with independents, and immigrant voters at D+22.

The two chambers start from different evidence. Senate races start from their own weighted polling average (recency with a 14-day half-life, sample size, pollster rating, house-effect correction), and each row shows the raw average, how many polls went in and the most recent one. Poll correction (13 October): every Senate average is then moved 3.5 points toward the Republican, on the assumption that Senate polls understate Republicans by that much, in line with the misses of 2014 and 2020. It is an assumption, not a law: in 2022 Senate polls were close to right, so the direction of the miss has not been stable. Democratic gains since August (13 October): Harry Enten’s aggregate (CNN) has the Democrat up since August by 2 points in Iowa, 3 in Ohio and 5 each in Alaska, Kansas and Michigan, and those gains are added after the correction. Today’s averages already carry part of them (since 16 September ours have moved Alaska +2.7 and Michigan +0.7, Iowa and Ohio barely or backwards), so this partly double-counts. The random national miss in the simulation is centred on the adjusted figure. House districts are not polled regularly, so each starts from the generic congressional ballot (D+6.3, 30 August) plus twice its Cook PVI. Cook ratings (October): the Cook Political Report moved eleven House races toward the Democrats. A rating is scored at the middle of this page’s bands (toss-up 0, lean 4.5, likely 8 points). For the three we track, the starting margin is the average of the formula and Cook’s new rating, because Cook weighs incumbents and candidates and the formula does not: OH-01 Lean→Likely D, OH-10 and OH-15 Likely→Lean R. Cook still rates OH-10 and OH-15 Lean Republican, so they move toward the GOP here even though Cook moved them toward the Democrats; the formula had them closer than Cook does. The other eight seats join the simulation from Cook’s rating alone instead of being assumed to vote as in 2024: FL-25 Toss-up→Lean D, MT-01 Lean R→Toss-up, MN-01, NY-21 and SC-01 Likely→Lean R, and FL-16, KS-02 and NE-01 Solid→Likely R. We have no gas readings for those eight.

Known blind spots. The House formula ignores incumbents (only the Cook-rated seats below take them into account) — Brian Fitzpatrick (PA-01) and Mike Lawler (NY-17) have repeatedly outrun their districts’ lean. House control assumes the 407 unmodelled seats go as they did in 2024, which ignores the 2025 mid-decade redistricting. The Senate count assumes seats outside the fourteen tracked races stay with their current party. Nebraska’s Dan Osborn is an independent and is counted separately.

Sources. Polls: the ElectIndex poll file, which compiles published surveys; pollster names and dates are shown per race. Gas: AAA. House lean: Cook PVI.