The 2026 State Navigate forecasts build on the model we used in our 2025 Virginia and New Jersey forecasts. The basic process remains the same: use previous election results to estimate each district’s partisan lean, adjust those results to the expected election environment, combine them with Executive Director Chaz Nuttycombe’s ratings (which are largely built off of a formula but also rely on some qualitative analysis), and simulate a range of possible outcomes. This year’s changes expand that framework across states with different electoral histories, candidates, and political environments.
The foundation of the legislative forecast is still the district’s recent election results. We use a combination of presidential, gubernatorial, and state legislative elections, with the selection and weighting varying by state and chamber. Previous legislative results help capture district-level voting patterns that statewide elections can miss, including an incumbent’s ability to attract voters who usually support the other party, otherwise known as “personal votes” vs. “normal votes.”
Before combining those elections, we adjust each result relative to its statewide baseline. For example, if a Republican won a district by 10 points in an election where Republicans won statewide by four, that district performed six points more Republican than the state. If our projected statewide environment were a two-point Democratic advantage, that historical result would translate into a four-point Republican advantage in the district.
That calculation gives us a common starting point for comparing elections held in different environments. We also account for situations where a past result is less useful, such as an uncontested race or a seat where the incumbent has retired. A candidate winning without meaningful opposition should not make an otherwise competitive district look permanently safe.
One change for 2026 is that the balance between historical results and Chaz’s ratings is more flexible. Last year’s Virginia methodology described a roughly 85% election-data and 15% ratings split. This year, the starting weights vary by chamber, and the model adds weight to legislative results and Chaz’s ratings in districts where past election margins differ more substantially. This helps account for places where presidential voting alone does not tell the full story.
The statewide environment has also changed. In 2025, the governor’s race provided the central statewide projection for the legislative forecast. For 2026, we begin with the state’s underlying partisan lean and a national political-environment estimate. That national estimate combines generic-ballot polling—which asks voters which party they would support for Congress—with an estimate derived from presidential approval. The current blend gives 65% of the national signal to generic-ballot polling and 35% to the approval-based estimate.
Where a governor’s election is taking place, the legislative forecast can also incorporate the projected gubernatorial margin. Its influence varies by chamber. This allows the model to reflect both the national midterm environment and the possibility that a state’s gubernatorial candidates run ahead of or behind their parties.
Once we have a projected margin for each district, we simulate the election 25,000 times. These simulations account for both movement in the statewide environment before Election Day and uncertainty in individual district outcomes. The amount of possible movement generally decreases as the election approaches, but uncertainty remains even in the final forecast.
As in 2025, the simulations use electoral and demographic similarity to connect districts. Two districts with similar voting histories, racial composition, and educational attainment are more likely to move together than two very different districts. If we underestimate one party among a particular group of voters, that mistake may show up across several similar seats.
For 2026, that relationship also extends across the two legislative chambers. Similar House and Senate districts can move together, and both chambers share a statewide swing. We also connect chamber outcomes with the governor’s race when calculating the chances of unified or divided government. A favorable environment for one party can help it across multiple offices, while still allowing split-ticket outcomes.
In each simulation, we count the seats won by each party and determine whether the result produces a majority, a tie, or a supermajority under that chamber’s rules. Offices without an election remain fixed when calculating control of state government. This gives us both the odds in individual districts and the range of possible outcomes for the legislature as a whole.
The governor forecasts have received a more substantial update. They still combine polling with a fundamentals baseline, but the 2026 fundamentals model goes beyond the Virginia-specific relationship between presidential approval, economic conditions, and gubernatorial results described last year.
The new baseline uses historical gubernatorial elections to estimate how a race should look given the state’s partisan lean, the national environment, the incumbent situation, candidate experience, and fundraising. Open-seat races and races with a sitting governor use different equations. For incumbents, the model considers their previous electoral performance and how much they ran ahead of the broader partisan environment. Candidate experience is represented through prior offices held.
Fundraising becomes more influential as the cycle progresses. We blend versions of the fundamentals model with and without fundraising, giving the fundraising version more weight later in the year. When a reporting date is available, that weighting reflects the age of the financial information.
Polling then updates the baseline. Polls receive different weights based on recency, sample size, the population surveyed, and partisanship. The model also limits repeated influence from the same pollster. Races with little usable polling depend more heavily on fundamentals; races with more polling give surveys greater influence.
Independent and third-party candidates are modeled separately, including assumptions about their support and how they draw votes from the major parties. Their polling support is adjusted for possible decline before Election Day. The uncertainty calculation also recognizes that polling and fundamentals can make similar errors, rather than treating them as completely independent evidence.
The projected margin is our central estimate, while the win probability describes how often a candidate or party wins across the modeled range of outcomes. A 70% chance of winning does not mean 70% of the vote, and it still leaves a meaningful chance of losing. Our goal is to show both what the available evidence suggests and how much room remains for the election to turn out differently.