The 2026 polls fell 3.5 points short of the Democrats
The analysis published on November 13 by Strength In Numbers measures the cycle's average error and ranks it as the fourth largest since 1998. The failure was not in who answered the poll, but in the likely-voter screen: registered-voter numbers missed by 2.6 points and likely-voter numbers by 4.4.

Democrats beat the result given to them by the final three weeks of polling averages by 3.5 points on average, according to the analysis G. Elliott Morris published on November 13 in Strength In Numbers. The calculation covers the generic House ballot and the Senate, House and governor races with at least 95% of the vote counted. The typical error since 1998 is about 2.8 points in either direction, so 2026 stands as the fourth-largest miss in the series, behind 2020, 1998 and 2016. What is unusual is the direction: it is the first time since 2012 that polls underestimated Democrats.
The miss was even across offices: 3.3 points in the Senate, 3.3 in the House and 3.2 in governors' races. Democrats ran above their polls in 18 of the 22 Senate contests measured. The extremes were Georgia, where Jon Ossoff ran almost 10 points above the polling, and Texas and North Carolina, at about 9 points each. On the generic ballot the gap was even wider: final averages gave Democrats a lead of about 9 points and the party is on track to win the national House vote by around 14, roughly 5 points of underestimation.
The analysis points to a specific culprit, and it is not non-response. Over the final three weeks, public polls reported both registered-voter and likely-voter results 32 times; in 25 of those 32 cases the likely-voter number was more Republican. Across those 25, the likely-voter figure missed by 4.4 points on average and the registered-voter figure by only 2.6. Screens based on vote history moved results toward Republicans, and those of Marist and Fox News moved them the most, by 3 to 4 points on average. Fox's Texas poll illustrates the problem: Democrats were 6 points ahead among registered voters and tied among likely voters; they won by 12.
Turnout explains much of the rest. Morris calculates that the profile of those who actually voted moved the electorate about 6.5 points relative to 2024, while his own model had moved it only about 3. His final likely-voter sample had a turnout ratio of 1.06; filtering to respondents who said they would «definitely» vote raised the ratio to 1.14, which is what reality matched. New voters broke for Democrats roughly 75 to 25, and infrequent voters backed them by 16 points after supporting Donald Trump by 7 or 8 in 2024, according to Joy Wilke of BlueLabs. The reading is not new to pollsters on either side: on November 4, Chris Anderson and Daron Shaw, the Democratic and Republican partners on the Fox News poll, had already told that outlet the surveys «underestimated» the Democratic vote, and David Paleologos of Suffolk University summed it up: «we were low on the margins of victory».
When a cycle's error is concentrated in the likely-voter screen rather than the sample, the fix is not calling more people: it is revisiting the turnout assumption. Ask your pollster for both numbers, registered and likely, and demand the turnout ratio being assumed. In 2026 that ratio should have been 1.14 and almost everyone worked with 1.06; that difference was worth 3.5 points of margin.
Summary written by the FOCUS editors based on the original story. Rights to the original content belong to its publisher.
FOCUS courtesy space
Voter and consumer dataTrusted voter and consumer data at your fingertips.Over 50 years as a source of enhanced voter, consumer and constituent data, on a web platform.l2-data.com →

