What we got right and wrong in 2026

A forecast is only worth reading if its makers are willing to grade it in public. Ours was frozen on the morning of November 3, before any polls closed, and every file was fingerprinted so it cannot be quietly edited afterward. This piece goes through how that forecast held up against the count, using the same scores as the accuracy report. Some races are still being counted, so a few numbers below will move as the final certified results come in.

The headline: three chambers, and the national vote to a tenth

The forecast picked the winner in 476 of 506 Senate, House and governor races, counting leaders in races that were not finished. Among races that were fully counted when we scored them, it was right in 413 of 420, or 98.3 percent. It favored Democrats to win the Senate at 81 percent, the House at 91 percent and a majority of governorships at 91 percent, and all three came true.

The part we are proudest of is the national vote. Our final national survey of 1,170 registered voters, fielded October 28, had Democrats ahead by 13.1 points on the generic congressional ballot. The projected final national House vote is D +13.0. The major polling averages closed near D +8.5, which means the industry as a whole underestimated the Democratic wave by about 4.5 points. Of the 42 final national polls we tracked, ours finished closest.

The Senate median landed almost exactly on the result. The House is where we were too cautious: the wave carried Democrats past our median of 254 seats to about 262, still inside the range we published but toward its top end. The reason shows up in the race by race errors below.

The races we got wrong

Statewide, the forecast missed 4 races. Each one is worth a look, because they failed in different ways.

• Nebraska Senate. We made Dan Osborn the favorite at 59 percent with a projected margin of Osborn +2.0. Pete Ricketts won at R +6.5.
• Alaska Senate. We made Mary Peltola the favorite at 75 percent with a projected margin of D +5.1. The first round count, with 72 percent of the vote in, has Dan S. Sullivan ahead at R +0.9, and the ranked choice tabulation is still to run, so this one is not settled.
• Mississippi Senate. We made Cindy Hyde-Smith the favorite at 53 percent with a projected margin of R +0.5. With 97 percent counted the race is essentially tied, so this miss may yet reverse.
• Texas Governor. We made Greg Abbott the favorite at 52 percent with a projected margin of R +0.2. Gina Hinojosa won at D +7.0.

Mississippi and Texas went in as near coin flips, and a well calibrated forecast should expect a few of those to break the other way each cycle. What matters in Texas is the size of the miss, not the call: Gina Hinojosa outran even a forecast that already treated Texas as unusually sensitive to the national environment, and that error is part of a pattern described below. Nebraska was a clearer miss in the opposite direction. We leaned on Dan Osborn's strong 2024 showing as an independent, and the state returned to its partisan baseline more than the model expected. Alaska's count is not final until the ranked choice rounds are run.

Where the errors clustered

Across every race we scored, the typical miss on the margin was 4.1 points and the results came in 1.1 points more Democratic than forecast on average. 87 percent of results landed inside our 80 percent ranges, a little more than they should, which tells us the ranges were slightly too wide this year.

The county results are more revealing. We forecast all 3,192 counties in the statewide races, and the vote weighted miss was 5.0 points. Suburban counties, which cast the most votes, came in 2.5 points more Democratic than forecast. Rural counties were about even. The biggest regional miss was the Southwest, 5.0 points more Democratic than forecast, while the Lower Midwest and Plains went the other way, 2.3 points more Republican than forecast.

Put together, the story is consistent: Democrats gained most in the suburbs and in heavily Hispanic counties of the Southwest, more than our demographic swing estimated, while the Plains held closer to their usual lean. That is how the House median came in low, how Texas was missed, and how Nebraska was missed in the other direction.

Our own polls

On Point Polling also fielded state polls during the cycle, and they are scored separately so they cannot hide behind the forecast. In the 8 Senate races we polled, our final poll had the right leader in 7 and missed the margin by 2.7 points on average. The 8 governor races were harder: 6 right, an average miss of 5.3 points, with results 3.5 points more Democratic than forecast. We publish every one of these, misses included, on the accuracy report.

State legislatures

Below the statewide races, the forecast called 80 of the 83 legislative chambers that are settled and the winner in 5,331 of 5,641 individual districts. Legislative districts are rarely polled, so those calls rest almost entirely on fundamentals and the national environment, which makes them a useful test of the model's foundation.

What we take into 2028

The lesson is not that the polls were wrong. Our own national poll was nearly exact. The lesson is about translation: how a national wave spreads across suburbs, Hispanic communities and the rural Plains. For the 2028 forecast we have rebuilt the electorate model on vote history, splitting every county into groups by how often they vote and calibrating each state to its 2026 turnout, so that the next forecast starts from who actually showed up rather than from who usually does. We explain that model in who votes in 2028.