The EZ Dub Election Forecast

How to read it

• Click or tap a race for its numbers in a panel on the right; What's driving this prediction? opens the same explainer as the forecast page: every candidate, the markets and the factors behind the numbers, each defined. On the district map the first tap zooms to the state.
• Four views: Margin shades each race by the model's projected margin of victory in points, Win % by the favored party's chance to win (the model's or a market's), Win % model vs market by how far the model's win chance is from Kalshi or Polymarket, and Margin vs markets by how far the model's margin is from Polymarket's.
• Market margins: on the Margin view, Kalshi shades each race by its average margin, the probability-weighted result implied by its margin brackets ("Talarico by 2-4%", straight bets on the final margin in 16 headline races) or else its "Democrats win by at least N points" markets; Polymarket by the median of its bracket ladder ("Democrat by 3-6%", normalized to 100%). Most races list only the favorite's side; a race stays gray when the markets don't price enough of the outcome to read.
• Margin: the margin model's projected lead of the front-runner over the runner-up, in points; deep blue or red is a margin of 30+ points, light is a close race, gold is under 1 point. The margin comes from a separate model from the win chance, so in a few close races the two pick different leaders ("Models split"). Where the top two are from the same party (California's top-two, Louisiana's all-party first round) the shade is that party's color and the margin is between the two of them.
• Margin vs markets: each race is shaded by the margin model's projected result minus Kalshi's average margin or Polymarket's ladder median (switch with the venue buttons), both signed toward the Democrat, in points. Blue means the model expects a bluer result than the market, red a redder one, pale within 2 points. Races with no ladder median or no model margin stay gray.
• Win % model vs market: each race card and table row names our model's favorite and gives that side's chance from our model, Kalshi and Polymarket side by side (an independent in the Democrats' place, like Nebraska's Osborn, counts as the Democrat). The shading is signed toward the Democrat: our model's chance for the Democrat minus the market's, blue where the model gives the Democrat a higher chance, red a lower one, pale within 2 points. The table lists the biggest disagreements first.
• Our model: each race's shade is the model's chance that its favored party wins the seat (that party's candidates combined), from light at 51% to deep near 100%; gold means the favorite is under 51%. The labels are the forecast page's tiers (Safe 95%+, Likely 80-95%, Lean 60-80%, Tossup under 60%); the forecast page rates each candidate and explains every race.
• No polls: races no pollster has covered, most House seats, use the model's no-polls version: which party holds the seat, its past results (the new map's lean where districts were redrawn for 2026), the national environment, fundraising and the candidates' experience. Tick Polled races only to gray them out.
• Kalshi and Polymarket: the same tiers applied to each venue's price for each party to win the seat. Prices are normalized so the outcomes add up to 100% (the venue's built-in margin removed), from the dashboard's last market refresh. Gray races have no party market on that venue.
• House map: the 435 districts on the lines in force for November 3, 2026. Nine states redrew since 2024 (Texas, California, North Carolina, Ohio, Utah, Florida, Tennessee, Louisiana and Alabama); Missouri and Virginia vote on their old maps after courts blocked the new ones. Click a state to zoom in, use + and −, and drag to move around. Squares shows each state's districts as equal squares instead, so small urban districts are as easy to see as rural ones.
• District profiles (in each district's card): the Census Bureau's 2020-2024 American Community Survey. For the 173 redrawn districts there are no official figures yet, so they are summed from block groups on the new lines; the same method reproduces the official figures for the old lines to within about 0.1 point. Redrawn districts also show the same-numbered district on the 2024 lines.
• What's driving this prediction: click a race, then the button in its panel (or Why beside any race in the table). It opens with the inputs that move the model most in that race: which candidate each one helps and how much, both candidates' values, and what it measures, then every candidate's chance and margin and the markets. The forecast page lists every input the models use.
• The model updates with every model run and the venues with every market refresh. Shapes: the Census Bureau's cartographic boundaries (states via us-atlas; 119th Congress districts) and each redrawn state's official 2026 plan.

Every number on this page