The counts are not finished. California, Washington, Oregon, Colorado and Alaska still have a large share of their vote to report, and the figures below will move. But in the races that are nearly fully counted, the pattern is already clear: the final polling averages understated Democratic candidates.

We compared each race's polling average on 2 November, the one our final forecast used, with the current count. The margin is the Democratic candidate's share minus the Republican's (the independent's, where the independent was the main opponent). A positive number means the polls underestimated the Democratic candidate.

The size of the miss

Races with at least 95% of the expected vote counted:

	Polled races	Polls underestimated Democrats by, on average	Median	Races where polls underestimated Democrats
House	77	+4.5 points	+4.1	65 of 77
Senate	22	+2.7 points	+2.9	16 of 22
Governor	18	+1.6 points	+2.3	13 of 18
National House vote	poll average D+8.7, count D+10.3	+1.6 points

The national row compares the final generic-ballot average with the House vote counted so far in districts with both a Democrat and a Republican on the ballot; it is provisional until the West finishes counting.

Races with three serious candidates (Montana's Senate race, for example) are left out, because a two-candidate margin does not describe them.

A two- or three-point error in a single race is ordinary. What makes this one notable is its direction. Polling errors in a year are correlated, and this year they ran the same way in most of the country, in Senate races as well as House races, in Texas and Georgia as well as New Hampshire and Minnesota.

Some of the largest Senate misses were in the South: Texas (polls D+4, count D+12), Georgia (D+9 against D+18) and North Carolina (D+12 against D+20). Mississippi, still uncalled, is the sharpest case of thin polling: three polls all year, all sponsored by Democratic-aligned groups and the last on 22 October, averaged Republicans +7. The count is now within a few dozen votes. Among governors, Georgia, Texas and Nevada each moved by eight points or more.

Where it went the other way

The error was not uniform. In several of the most Republican states the polls overstated the Democratic or independent challenger: South Dakota's Senate race (10 points), Idaho's (8 points), and the South Dakota and Idaho governor races. In Vermont, polls showed a close governor's race, and Phil Scott, the Republican incumbent, won by about 7 points. Three Georgia House districts (the 3rd, 7th and 11th) also came in more Republican than their few polls.

The House polls and the generic ballot

District polls are sparse and many are sponsored by a campaign or party. We checked whether that explains the House miss. It does not: in districts where every poll was partisan, results still ran 3.4 points more Democratic than the average; in districts with at least one independent poll, 5.5.

The national question ("which party's candidate will you vote for in the House?") averaged Democrats +8.7 in the final week. Adding up the House votes counted so far gives Democrats +10.8 (+10.3 counting only districts with both a Democrat and a Republican on the ballot). That gap will change as the West finishes counting, so treat it as provisional.

What it did to our forecast

Our model did not use the polls raw. Based on 13 elections from 1998 to 2022, in which the generic-ballot question overstated Democrats on average, it moved that average 2.7 points toward Republicans before combining it with the race-by-race polls and the fundamentals. This year that correction went the wrong way, and it added to the polling error rather than offsetting it.

The seat counts show the effect:

	Forecast median	Forecast 80% range	Result so far
House, Democratic seats	238	222–258	234 called; Democrats lead in 28 of the 63 not yet called
Senate, Democratic seats	53	49–56	54 called; Democrats lead in 2 of the 2 not yet called
Governors, Democratic	22	18–25	22 called; Democrats lead in 2 of the 2 not yet called

The Senate and governor results fell inside the forecast range. The House result is on course to finish above it: in 79 of 248 nearly fully counted districts the Democratic margin beat the top of our 80% range, against 1 that fell below the bottom. That one-sidedness is what a shared polling error looks like in a model, and it is the main thing this forecast got wrong.

We will revisit these numbers when counting is complete and publish the race-by-race comparison.

Method: polling averages and forecast ranges are from the final forecast run on 2 November (archived). Counts are the live results as of 4 November, 20:06 UTC, from state election offices and CNN. "Expected vote counted" is CNN's estimate.