How the House Forecast Works
Designed by Logan Phillips
The Race to the WH House Forecast is an analytical data-driven forecast that predicts the outcome of every Congressional District in the nation using a combination of factors that include the recent voting history of each district, the incumbent’s last election, fundraising, and the latest polling. It simulates the election 10,000 times after every update to predict the chance both parties have of winning the majority.
The forecast is deeply rooted in electoral history, based on a mountain of evidence from past elections held over the last few decades. Every feature in the model has demonstrated predictive power in past cycles, and the overall forecast has been fine-tuned to be as accurate as possible.
On the website, I present the forecast’s findings with interactive graphics. My goal is to make it easy for people that aren’t following politics too closely to understand the shape of the election, while also providing detailed information for those that follow the race closely and want to dive deeper into the details.
Since we started forecasting House races in 2022, our track record has been one of the best in the nation. Of the nine major election forecasters that have forecasted the last two congressional cycles. Race to the WH has been the second most accurate next to FiveThirtyEight, calling 97.24% of races correctly. In 2022, we came just one seat short of perfectly predicting each party’s seat total.
That said, there’s always room to improve. In 2024, the forecast excelled at predicting individual races, but it overrated Democrats’ chances of winning the majority. This cycle, I’ve worked hard to improve our projection for the national popular vote, which I discuss in more detail in the National Political Environment section.
Broad Overview of How the Forecast Works
Before diving into each key component, here’s a broad overview of how the House Forecast works. The House Forecast has three primary objectives:
1. Predict the Margin of Victory for Every Race
2. Calculate the Chance of Winning Both Candidates Have in Each Race
3. Forecast the Chance Both Parties have of Winning the House Majority, and the # of Seats they are Most Likely to Win
When I calculate the Margin of Victory for each individual House race, I’m trying to predict the lead in percentage points that the leading candidate will have over their top opponent. For example, if the Democrat wins 50.1% of the vote and the Republican wins 49.5%, the margin of victory is D +0.6%.
For each race, the forecast uses up to seven factors to assess who’s favored. Collectively, these seven factors take into account the political geography of the district based on how it’s voted in recent elections, the latest polling for the race, and signals of candidate quality for both parties. The projection for each race updates daily based on shifts in the national environment and new race-level information.
Second, the Forecast calculates the Chance Both Candidates have of Winning the Race. No forecast can call every race correctly, and so a good election forecast doesn’t just predict the winner but also predicts the chances that an upset occurs.
Our forecast calculates how much we expect the actual results to deviate from the Projected Margin of Victory, based on what has happened in thousands of past Congressional elections used for back-testing. If there is a high chance of a miss in a race that is projected to be close, then there will be a very high chance of an upset.
Third, I feed the race-level projections into our Election Simulator to Forecast the Chance Both Parties have of Winning the House Majority. It runs through 10,000 simulations of the election for every update. It simulates a wide variety of scenarios, where parties over and underperform across different regions, key demographic groups, and states.
Part 1: Projecting the Margin of Victory
To predict the Margin of Victory in each House race, (or the lead in percentage points that the leading candidate will have over their top opponent), the forecast analyzes each race using up to seven different factors:
Partisan Lean of the District
Incumbent’s Margin of Victory in the Last Election
Experience Winning Elections
Fundraising
Polling Average
National Political Environment
Margin from the First Round of Elections (California and Washington Only)
1. Partisan Lean of the District:
Every District has a partisan lean, which is the net percentage that it has favored Democrats or Republicans relative to the national popular vote in the last two presidential elections. Let’s use Michigan’s 7th district as an example. Trump won the district narrowly in 2024 by 1.3% (49.9% to Harris’s 48.6%). However, Trump also won the national popular vote by 1.5%, by 0.2% more than he won MI-7. That congressional district voted D+0.2% relative to the national popular vote.
To calculate the Partisan Lean, we take 75% of its Partisan Lean comes from the 2024 election, and 25% comes from the 2020 election. This provides a strong starting point for how a district might vote in a neutral cycle if both candidates are equally strong
2. Incumbent's Margin of Victory in the Last Election
One of the best ways to predict the results of the next election is to use a Representative’s performance in their last election. This isn’t done blindly – each election has unique dynamics that must be corrected for to make the data useful.
For example, let’s consider a Democratic Congresswoman who won her seat by 2% in 2008. Without context, her performance might indicate she’d be a mild favorite to win in 2010. However, 2008 was one of the best years for Democrats in generations, and they won the House Popular vote* by 9.1%. So that 2% victory would imply a Republican would win the district by 7.1% if 2010 were a neutral election.
In 2024, the Republican Party won House elections by 2.26% nationally*, so we adjust every Incumbent’s margin of victory towards Democrats by 2.26% for 2026. We also adjustments for districts whose partisan lean changed between 2024 and 2026 after the last Presidential election – including for candidates that are now running on entirely different turf thanks to the wave of states redrawing their districts midway through the decade to give one party an advantage.
Finally, we make an adjustment for the type of election the incumbent ran in last cycle. It’s far harder for a challenger to beat an incumbent than for an incumbent to win re-election, so we give challengers a significant boost in their last margin of victory to predict how they will do in the next election. We also give modest boosts to members that won an open district.
* Our “national popular vote” differs from the number you’ll see published elsewhere, which simply uses the total number of votes for Democrats and Republicans. The standard number is misleading, especially in midterms, because there are some districts where the parties are unable to recruit a single candidate to run. In those districts, Democrats or Republicans may win 100% of the vote, which distort the national picture. As a result, we calculate the true national vote by correcting for Congressional districts where one party ran unopposed or only faced a third-party challenger. We project how those districts likely would have voted if it had to decide between a Democrat and Republican, based on its voting patterns in recent elections, and the national environment in that election cycle.
3. Experience Winning Elections
Decades of past election cycles show us that candidates with a history of winning elections perform better on average than first time candidates.
In each race, we identify the experience level of the candidates. Each type of experience is assigned a point total, and the candidate with the higher point total gets a boost. Incumbents perform much better in elections than first-time candidates, state senators, local politicians, and state reps. They do, however, underperform against governors, senators and big-city mayors. If we only have a likely nominee for one party, the impact of Political Experience is cut in half, and we assume their opponent will have the same level of experience as the average non-incumbent over the last decade.
4. Fundraising
The election forecast looks for every sign that a candidate might be uniquely strong and positioned to overperform expectations – and one of the most useful indicators their ability to fundraise. Strong fundraising suggests a candidate has a serious operation, the ability to mobilize grassroots supporters, and that they’ll have the resources they need to get their message out.
To factor Fundraising into the forecast, I compare the total money raised from individual donors by each candidate - which excludes money from PACs and self-funding. The higher a candidate’s share of the combined total, the larger the boost they receive in the forecast.
Additionally, I give a boost to challengers running against incumbent House members, who have an early natural edge in fundraising due to their existing contacts, email lists, and support systems. This boost is far greater early in the cycle, when most challengers trail incumbent, but it tapers as we get closer to the election.
Fundraising is only a factor when we have identified a likely nominee for both parties, and when both candidates have been active long enough in the last financial quarter to have a fair opportunity to raise money.
5. Polling Average
The fifth factor is the Polling Average. While polling has been less accurate in recent elections cycles, it’s still one of the best signals to predict the result of an election, particularly as we get closer to election day. Polling’s weight is limited early on, but in a well-polled House race late in the cycle, it becomes the most important factor.
As with the Senate and Governor races, our Polling Average gives more weight to recent polls, to pollsters with excellent track records, and to those with larger sample sizes. We also correct for bias.
In House races, a large share of the House polls come from partisan pollsters aligned with one candidate. We are very aggressive about correcting for bias with these pollsters, and they receive half as much weight in the average as non-Partisan pollsters.
The total weight given to polls in the Forecast differs for each race based on the quantity and quality of the polling. The maximum impact it can have is capped early on in the race when polling is less predictive, but it can make up the majority of the forecast in the final months of the race.
6. National Political Environment
House races don’t exist in a vacuum – and from cycle to cycle, parties tend to do better or worse across the board based on the National Political Environment. This is one of the hardest parts of the election to predict, especially in an era when polling has become significantly more challenging. Voters are far less likely to respond to phone calls from strangers or spend time on surveys, which requires pollsters to be more creative in how they reach the public and ensure their samples reflect the electorate. Considering the challenge they face, I would argue that pollsters have actually performed quite well over the last four years.
Lower response rates mean there is inevitably a greater risk of big polling misses, so we’ve had to go beyond polling to estimate the national environment. After extensive testing, we’ve settled on three inputs to forecast the combined national House vote in midterms - a framework that has performed very well in back-testing across past decades.
First, in midterm elections, we start with a baseline favoring the opposition party to the President. The American people have a long history of voting for checks and balances on the sitting President in midterms, particularly in the last half-century. In every election but one since 1978, the opposition party has won the popular vote in House races, on average by 4.43%. The historic midterm result is our starting point for predicting the national environment, and the largest factor in our projection early on when the cycle’s unique dynamics are far from clear.
Second, we assess how each party is performing in elections held since the last election cycle compared to the partisan lean of the district that they are running in. More weight is given to recent elections, and to elections with larger voting bases. For example, a congressional district is weighed more than a state-house district.
We’ve found that historically, this has been even more accurate than the polling average. has been even more accurate than the polling average. However, there’s reason for caution. In recent years, Democrats’ coalition is now disproportionately made up of highly motivated college-educated voters that vote at higher rates in special elections. In 2022 and 2024, Democrats performed much better in special elections than they did in the November election.
Given how much special elections improve the forecast historically, it would likely be a mistake to outright exclude them. However, the older method used in 2022 and 2024 that utilized purely special elections needed refinement to avoid over-rating Democrats.
I delayed publishing the House Forecast by a month as I built a new version of our Special Elections tester from the ground up and tested it on each cycle going back to 2002. The updated method improves accuracy, especially in the most recent cycles. It now includes the California and Washington non-partisan primaries, where all candidates run in the same congressional contest and Democrats’ disproportionate turnout advantage typically doesn’t carry over. It also factors in results from the regular election day in November 2025. I found this has considerably improved accuracy across the last three election cycles, negated Democrats’ special-election edge, and almost perfectly predicted the outcome in 2022.
I also review each individual special election, and on rare occasions exclude races that will clearly not be predictive. In this election cycle, we had a New York State Senate special election in a predominantly Orthodox Jewish district whose partisan lead was R +51%. In the special election, it voted for the Democrat by a net +86%, a stunning overperformance far beyond anything else this cycle. Most voters in this district vote in blocs at the direction of influential rabbis who endorse candidates they feel best represent their communities; those leaders endorsed both Donald Trump and the Democratic state senator. This election was a unique case and had little predictive value for the 2026 election.
Third, I incorporate the generic ballot polling average. This polling average (which you can find here) aggregates polls asking whether voters prefer a generic Democrat or a generic Republican. polls which ask voters if they would prefer a generic democrat or a generic republican. It begins to have predictive value around mid-August of the year before the election and becomes considerably more accurate as we get closer to the election.
All three factors are combined into our projection for the National Vote. The exact weight of each factor shifts based on the phase of the election. The Generic Ballot accounts for about 25% early on and rises to as much as 75% in the final weeks, as history shows its predictive value improves dramatically late in the cycle.
The non-polling component is split between the Elections Results, and the Historic Midterm Average. Early on, the Historic Midterm Average carries far more weight, however Election Results gain more weight after the California and Washington Non-Partisan Primary Elections happen.
The Forecast adds a portion of the Projected National Vote to every House Race. A higher percentage is added to races where there is no incumbent.
7. Margin from the First Round of Elections
California and Washington have adopted open primaries in which every candidate runs in the same primary, and the top two candidates regardless of party make it to the general election. This provides us with incredibly valuable information about the preference a district has for Democrats or Republicans. While the result in November will differ to some degree from the first round due to changes in voter preferences and a different electorate, this remains a strong datapoint that considerably improves the accuracy of the forecast.
We exclude districts where something irregular makes the results unlikely to be predictive – like when the Green party takes a huge slice of what was likely going to be votes for Democratic candidates.
Combining this into the Projected Margin of Victory
We combine all the factors that are active in each race to produce the Projected Margin of Victory. When the forecast launches, the projection for most races is dominated by Partisan Leand the Incumbent’s Performance in the Last Election. At launch, the projection for most races is dominated by Partisan Lean and the Incumbent’s Margin of Victory in the Last Election. These are the two most heavily weighted non-polling factors (outside California and Washington) and the only components we can calculate from day one.
Later in the cycle, Political Experience gains weight once we can identify the likely nominees for both parties. Toward the end of 2025, we’ll begin factoring in Fundraising for a handful of House races in which both candidates have been running most of the year; that expands to dozens more in early 2026 and to nearly all races by midway through the year.
Polling Average becomes increasingly important in many top swing races as high-quality non-partisan polling accumulates. Additionally, we’ll see larger updates for California and Washington after their top-two primaries on June 2 and August 4, respectively; results are added on a district-by-district basis as states finish counting, a process that can take longer in California.
Part 2: The Chance Both Candidates have of Winning the Election
Any election forecaster worth their salt knows that they won’t always get it right. Once the forecast projects what will happen, it must also estimate how likely it is to be wrong, so that it can turn the projections into probability. The chance the forecast gives a candidate to win depends not just on the projected lead, but also on the uncertainty the forecast has around the prediction.
To forecast the chance to win, I calculate two things:
1. The Projected Margin of Victory
2. The Expected Deviation for Each Race
Expected Deviation measures how much we expect the Projected Margin of Victory to differ from the actual result. Thanks to testing the forecast on thousands of past elections, we can estimate how much, on average, we can expect the Projected Margin of Victory will deviate from the result based on what has happened in similar races past Congressional Elections.
The more information we have about a race, the smaller the expected deviation becomes. For example, if we have a House race with extensive polling and an incumbent that is running for re-election, and election day is just one day away, the projection is likely to be much more accurate. That’s because the forecast can rely on more tools (like the Incumbent’s Last Election and the Polling Average), and we don’t have to account for time left in the race where one side could surge ahead. In contrast, accuracy is going to be far lower when the election is more than a year away, it’s an open seat, and there is zero polling.
These are the five factors that change the expected deviation for each race:
Is the Incumbent Running for Re-Election – Lower Deviation with an Incumbent
Days Until the Election – Lower Deviation Closer to the Election
Quality and Quantity of Polling – Lower deviation for races with multiple recent high-quality polls
Polling - Share of Undecided Voters – Lower Deviation with Lower # of Undecided Voters
Gap Between the Polling Average, and the Rest of the Forecast – Lower Deviation if the Polling and Forecast suggest similar outcomes.
After calculating the Expected Deviation and the Projected Margin of Victory, the Forecast converts them into a win probability using a normal distribution formula. This models how frequently Democrats or Republicans will end up leading the race.
In a race where the election is projected to be close, and the expected deviation is high, the chance of an upset will be high, and the race will likely be classified as a Tossup. However, if Democrats are projected to win by 25%, Democrats’ chances of winning the race will be well over 95% even if the expected deviation is extremely high.
A small portion of the projected win percentage incorporates ratings from Crystal Ball's and the Cook Political Report to capture qualitative factors that our quantitative model might miss – like a particularly charismatic candidate that’s a perfect match for the district, or a scandal-plagued candidate with a history of problematic comments.
Part 3: Simulating the Race
The last key piece is estimating how likely each party is to secure a Congressional Majority. The House Simulation runs all 435 districts 10,000 times per update to determine how likely each party is to secure the 218 districts they need to win.
These races do not exist in a vacuum. In every cycle, parties will outperform expectations in different ways that will change the map – ranging from the national vote, performances across regions, and across demographic groups. We simulate a wide range of scenarios to estimate how likely each party is to win a majority. Races that share similar traits are modeled with higher correlation. We determine the correlation of each race using:
Region
State
Racial Demographics
Incumbent Status
Partisan Lean
In every simulation, Democrats and Republicans will randomly over and underperform across certain kinds of races, to simulate how the election result can change based on what happens on election day. We then track how often each party wins across all simulations, along with the average number of seats they won.
This is calculated by using thousands of randomly generated numbers that shift the national environment, the unique dynamics of each individual race, how different racial groups vote, how parties do in different regions and states, and even if the parties perform particularly well in deep blue or deep red areas.
This matters so much because political trends tend to carry beyond the borders of each individual congressional race. For example, in the last two Congressional elections, overperformance in the right states changed the winner of the majority. In 2022, Republicans underperformed expectations nationally, but they were able to narrowly win the House based on their excellent performance in California and New York. Likewise, in 2024, Republicans were unable to replicate their performance in those two states, but they still secured the majority thanks to strong finish in Pennsylvania, where they won PA-7, PA-8, and PA-10 by a combined 15,447. If Democrats did just slightly better in the state, they would have won the House.
Part 4: How We Handle Redistricting
This year, President Trump has pushed Republican-controlled states like Texas and Missouri to redraw their maps to maximize Republican advantage in red-state Congressional races. Democrats in California are attempting to do the same. If they win a statewide referendum in November 2025 (which polls suggest is highly likely), their proposed maps will largely cancel out the gains Republicans made in Texas. Democrats also caught a break in Utah, where a judge found the Republican-friendly map violated the state constitution; the new map will likely turn a safe Republican seat into a Tossup.
For 2026, the new maps collectively will likely net Republicans 2–4 additional Congressional seats beyond what they otherwise would have won. That might be enough for Republicans to take the House if Democrats win the national vote by a small percentage.
Because I’ve forecasted the 2022 midterm elections, I forecasted the 2022 midterms—when every state with more than one seat had to draw a new map, this is an easy change to make. The forecast calculates the partisan lean of the new Congressional district based on how they voted in the last two Presidential elections, and it also adjusts the Incumbent’s Margin of Victory in the Last Election.
Changes to the Congressional Map (Last Change on October 7th):
We’ve added new maps for California, Missouri, Texas, and Utah. The maps for California and Utah aren’t official yet. In California, Democrats in the State Legislature have proposed a new map that would take effect if voters approve a state constitutional amendment in November. Polling suggests it will pass with ease. In Utah, the State Supreme Court struck down the previous map for partisan gerrymandering, and the State Senate has approved a replacement that could make two formerly safe Republican districts competitive. This is the most likely new map for Utah, though there’s still a chance the courts could reject it.