Senate
The race for control of the Senate is competitive, but clearly tilting blue; Democrats are favored to seize control of the chamber with a forecasted pickup of six seats, despite the historically unfriendly map. Here’s how it breaks down:
Here’s a table of our outputs, with probabilities and modeled margins. You can download the forecast by clicking “Get the Data”:
Here are the probabilities both parties have of winning control of the chamber:
Here are our projected seat counts:
Here’s how likely each scenario is:
House
The race for control of the House leans pretty strongly Democratic this year, with polling showing a blue wave amid Donald Trump’s cratering approvals. Here’s what the map looks like:
Here’s a table of our outputs, with probabilities and modeled margins. You can download the forecast by clicking “Get the Data”:
Here are the probabilities both parties have of winning control of the chamber
Here are our projected seat counts:
Here’s how likely each scenario is:
What about a polling error?
Polling error is a reality in every election, but it's hard to tell which side the polls will overestimate. In 2016, 2020 and 2024, polls overshot Democrats. In 2018 and 2022, they were dead-on — and in some cases, overshot Republicans. And in 2025, they substantially underestimated Democrats. Here's how the chambers shake out in our model in a series of different national environments.
Model Methodology:
First, we constructed a generic ballot average of registered voters, which was weighted for pollster quality according to VoteHub’s pollster scorecard. Next, we calibrated the microdata from all of The Argument/Verasight’s monthly polls in 2026 to this baseline average and then applied a likely voter screen to the survey microdata.
After doing this, we calculated the swing from the presidential margins at a state, regional, and national level, for each demographic — with each geography’s swing weighted for sample size in the composite aggregate. We then constructed a likely electorate at the district and state level alike and applied this demographic-level swing to that electorate, before adjusting each estimate for incumbency, candidate fundraising, outside spending, and prior candidate overperformance (using Split Ticket’s Wins Above Replacement scores).
This was then blended with two other components: a statistical model trained on the last four election cycles and polling averages internally computed for each House and Senate race. Finally, we conducted a Monte Carlo simulation, with correlated error terms on demographic, state, and national levels.
Aggregated polls were gathered via VoteHub’s polling API. Polling microdata was taken from monthly The Argument/Verasight surveys. Likely voter screens were created through a mixture of data from the Catalist voter file and self-reported voter enthusiasm.