After several months of development, The Data Times U.S. Midterms Forecast is finally here! On this page, you’ll find the ratings, vote share projections, and probabilities for control of Congress. You’ll also be able to view the in-depth breakdown for the projections of all seats up for election this year. In total, we’re forecasting 470 races across the country.
Jump here to skip to the Senate, and here to view our methodology,
The Democratic Party
is currently favored to win the
2026 U.S. Midterm Elections
Nationwide House Projections
Vote Share Projections by Party
District Heatmap
Seat Projections by Party
Gerrymandering Effect: R +1.35%
This means that Democrats need to win the nationwide popular vote by roughly 1.4 points to secure a majority of the chamber’s seats.
U.S. Senate Projections
Methodology
Note: This is a brief explanation of the numbers shown on this page. For more detailed information, contact inquiries@thedatatimes.com
The TDT Midterms Forecast includes the most complex district- and state-based models we’ve made to date, accompanied by large advancements to our polling averages and variable weighting. The core element of the model (both for the House and Senate) is the national Generic Congressional Ballot projection. This projection uses a dynamic recency-scoring system, ensuring the model has ample polling data to work with in any given time period. For more sparse polling periods, the model relies on older polls to balance out the average. This methodology also extends to our Senate polling, though to a lesser extent. During periods when many polls are being released, the model automatically filters out older surveys as long as specific scoring thresholds are met. During sparser polling periods, the model permits these same surveys. Alongside our recency-scoring for polls, we’ve expanded our pollster grading library to accommodate nearly 90 surveying firms. Lastly, our sample scoring now automatically weights sample sizes based on the groups surveyed.
Surveys that publish GCB results are weighted based on the factors listed above, resulting in a final projection for the nationwide popular vote. Accompanying this projection is our Margin of Error formula, which takes into account variables such as undecided voters, overall survey scores, published MoE data for each poll, specific sample sizes, and the time until election day. Our final calculated Margin of Error aids in the determination of deviations for both individual seat probabilities and overall House control. Furthermore, the raw GCB projection is modified (usually with a reduction) based on the calculated MoE for the purpose of seat projections.
Once the GCB and MoE are calculated, the model determines the relative swing from the 2024 national popular vote (NPV) to the current projections for November. This calculation is then used to apply a proportional swing to every district from its 2024 congressional results. For states that have redistricted since 2024 or for districts that do not have incumbents seeking reelection, the model uses adjusted presidential data as a baseline for the swing. Alongside the proportional swing is a nominal shift in percentage points, reflecting the fact that not every district will swing proportionally (as levels of support are more saturated in some districts than others).
District projections are also calculated using an application of the GCB to the district’s partisan leanings. This variable helps ground the model in some districts while also increasing the shift in others to help account for demographic baselines. The partisanship of a district is based on how its recent voting history differed from that of the nation as a whole. Depending on the state and recent redistricting efforts, these partisan leanings are sourced either from the Cook Partisan Voting Index or calculated by The Data Times internally.
The last primary variable for the House is the district-level polling. These polls are weighted much more simply than the GCB polls, mainly taking into account overall recency and whether or not the firms are sponsored by a party/candidate. Considering Polling data is the only time-sensitive metric out of the four, it’s also the only one that’s weighted dynamically. It’s important to note that district-level polling is not an average of all polls conducted, but rather an aggregate of the most recent batch. This is because small differences in polling data usually do not have a significant impact on the projections for a given district.
The odds of a party winning in each district are calculated using a standard distribution curve. They’ll naturally increase as the election approaches because they use the baseline GCB-MoE formula. The Senate uses a similar foundation for its projections, though the core variables are weighted differently (and usually less than state-level polling). Additionally, the polls in each state contribute to unique, calculated MoEs for every Senate race that includes them. This is taken into account for the projected odds in each state.
The Senate is also more lenient than the GCB when it comes to older surveying, permitting polls if they’re either from a high-quality pollster or if there is a lack of polling in the state. This balance is calculated using what we call the “PollRec” score, a binary indication of whether or not the aforementioned state and surveying conditions outweigh the date of a given poll. For any state with five or fewer polls in the model, older firms can remain influential as long as they have been released in the past six weeks.
The final forecast for control of the U.S. House is based on the odds of the leading party winning the projected tipping point seat. This seat is determined using the structural bias of the districts nationwide (currently favoring Republicans due to recent redistricting) combined with the GCB projections at the time of modeling. Essentially, this means that the odds for each party winning control are determined by the likelihood of their final NPV margin either overcoming or taking advantage of the national district bias.
The control of the Senate is determined using a similar methodology. However, it differs in one key element. Rather than being calculated simply by the odds of winning the tipping-point seat with an increased MoE, it’s calculated by the odds of the leading party winning both the tipping-point seat and what we call the “cushion seat,” added to a small fraction of the odds that the leading party wins the other seats they’re currently ahead in without winning the more favorable ones. The “cushion seat” is the name given to the state that would be projected to flip right before the actual tipping-point state. In doing this, we add a bit more uncertainty into the model in anticipation of a potential polling error.
TDT Official Ratings
House: Likely D
Senate: Lean D
Hunter 67
