The Undecided Fifth What Pew's last big poll before the 2026 midterms really measured
Pew's last major survey before the 2026 midterms gave Democratic House candidates a 44% to 36% edge among registered voters. The headline was the eight points. The more telling number sat in the column on the right: one voter in five chose another candidate or was not sure, even after being asked which way they leaned. And they were not spread evenly.
An analysis of Pew Research Center data. Survey fielded Sept. 28 – Oct. 4, 2026.
- Democratic candidate
- 44%
- all registered voters
- Republican candidate
- 36%
- all registered voters
- Another candidate or not sure
- 20%
- 5% other, 15% not sure
- Consolidation gap
- 9 pts
- 82% of the Democratic side backed its party; 73% of the Republican side did
A note on timing, before anything else
This survey was taken a month before Election Day, among registered voters, not likely voters. The election was held on November 3. This piece is about what the survey measured and how to read it. It is not an account of the results, and a registered-voter poll in early October should not be read as one.
Each interactive figure awards a badge the first time you use it. There are six.
1 · The whole chart
The full picture, group by group
Here is the survey's central chart, rebuilt. Blue bars run toward the Democratic candidate, red toward the Republican, and the gold column is everyone who chose another candidate or was not sure. Tap a row to read it.
House vote preference among registered voters
- Democratic candidate
- Republican candidate
- Other / not sure
All registered voters: 44% for the Democratic candidate, 36% for the Republican, 20% other or not sure. Margin D+8.
Gender
Race and ethnicity
Age
Education
Party
2 · Where the lead came from
Subtract one side from the other
Take the Republican share from the Democratic share in each group and the survey's structure is easier to see.
Democratic margin by group, in points
Democratic share minus Republican share. Party rows are left out because they are near-tautological.
- Black+63
- Asian*+40
- Ages 18–29+28
- Postgrad+28
- Hispanic+27
- Ind/other, incl. leaners+16
- College grad+15
- Women+14
- 30–49+12
- All registered voters+8
- Some college+4
- 65++3
- Men0
- HS or less−4
- 50–64−6
- White−7
The Democratic advantage was widest among Black voters (63 points), Asian voters (40), voters under 30 (28), postgraduates (28) and Hispanic voters (27). Republican candidates led in three groups: White voters by 7, voters aged 50 to 64 by 6, and voters with a high school education or less by 4. Three patterns stand out, and each cuts against a simple story.
The gender gap was one-sided
Men split exactly evenly, 42% to 42%. Women favoured Democratic candidates by 14 points, but were also far more likely to be uncommitted: 24% against 16% of men. The gap came less from women being more Democratic than men (45% against 42%) than from fewer women choosing the Republican candidate (31% against 42%).
Age did not run in a straight line
Voters under 30 leaned Democratic by 28 points and voters aged 50 to 64 leaned Republican by 6. Yet the oldest group, 65 and over, tilted back to Democratic candidates by 3. The familiar assumption that support moves steadily rightward with age does not describe this data.
Education sorted voters cleanly
The Democratic margin climbed at every step: −4 among those with a high school education or less, +4 with some college, +15 among college graduates, +28 among postgraduates. Only race and party sorted voters more sharply.
3 · The undecided
The number that mattered more than the margin
And the uncommitted were not a random slice of the electorate. They were concentrated in specific places.
Share choosing another candidate or not sure
- Lean Rep38%
- Ind/other, incl. leaners36%
- Asian*27%
- 30–4926%
- HS or less26%
- Hispanic25%
- Ages 18–2925%
- Lean Dem25%
- Women24%
- Black23%
- Rep/Lean Rep21%
- All registered voters20%
- Some college20%
- College grad19%
- White18%
- Men16%
- 50–6416%
- 65+15%
- Dem/Lean Dem15%
- Postgrad12%
- Republican12%
- Democrat10%
Among the least committed groups were Republican-leaning independents (38%), independents and other-party identifiers, leaners included (36%), Asian voters (27%), voters with a high school education or less (26%) and voters aged 30 to 49 (26%). The most committed were self-identified Democrats (10%), Republicans (12%) and postgraduates (12%).
That profile matters for a practical reason. Younger voters and voters without college degrees are, historically, the groups least likely to turn out in midterms, and independents who lean to neither party are the least likely to say they voted. A share of that gold column was never going to become votes for anyone. A registered-voter survey cannot tell you which share.
How lopsided would the undecided have to break?
Pure arithmetic, not a forecast. Suppose all 15% who said “not sure” picked one of the two parties, and the 5% already choosing someone else stayed put. Drag to set how many of the not-sure voters go Republican.
Democratic 51.5 · Republican 43.5Lead: D+8.0
The lead disappears only when about 77% of the not-sure voters break Republican.
Undecided voters rarely break that lopsidedly, and many of them do not vote at all. That is why the 20% matters, and also why it cannot be read as a hidden Republican vote.
4 · Each side's own voters
A consolidation gap, not just a persuasion gap
The party rows at the bottom of the chart hold what may be the survey's most important structural finding: each party's own side was not equally locked in.
- Republican1284
- Lean Rep93853
- Lean Dem7225
- Democrat8810
- Democratic candidate
- Other / not sure
- Republican candidate
Among Democrats and Democratic leaners, 82% backed the Democratic candidate. Among Republicans and Republican leaners, 73% backed the Republican. Almost nobody crossed over in either direction: 6% of the Republican side chose a Democrat, 2% of the Democratic side chose a Republican.
The difference was in the leaners. Only 53% of Republican-leaning independents said they would vote Republican, with 38% uncommitted. Among Democratic-leaning independents, 72% were on board and 25% uncommitted. The eight-point lead owed at least as much to soft Republican-side voters sitting on the fence as it did to anyone changing sides.
Why the distinction matters
Voters who have switched parties are hard to win back. Voters parked in “not sure” who lean your way are, in most cycles, the easiest voters to bring home late. A lead built on the second kind is real, but more fragile than one built on the first.
5 · Engagement
The engagement picture had flipped since 2022
The same survey asked how much the election mattered to voters. On all three of Pew's measures, supporters of Democratic candidates were more engaged than supporters of Republican candidates — a reversal of the pattern Pew recorded in October 2022. Switch years to watch it turn.
- Supporters of Democratic candidates
- Supporters of Republican candidates
- “Really matters” who winsGap D+978%69%
- Thought “a lot” about the electionGap D+1450%36%
- “Extremely motivated” to voteGap D+1763%46%
The backdrop
- of all U.S. adults satisfied with the way things are going
- 22%
- of registered voters call the economy very important to their vote
- 84%
- say the president is a factor in their vote: 44% against him, 19% for him
- 63%
- presidential approval among all U.S. adults
- 30%
Midterms have often worked against the party holding the White House, and these numbers fit that long-running pattern rather than breaking it.
6 · A thought experiment
What if only one group voted?
“If only women voted” maps are a long-standing device in data journalism. Nobody is proposing to restrict the vote; the point is to see how much the shape of the electorate decides the outcome. Pew's survey is national and has no state-level numbers, so the map below is a simulation with one stated assumption, not survey data.
How the model works, and where it breaks
Method: each state starts from its certified 2024 presidential margin. The whole map is then shifted by the difference between a group's national margin in the Pew survey and the 2024 national margin (R+1.5). This is a uniform swing.
Where it breaks: groups do not vote the same way everywhere. White voters in Vermont and White voters in Mississippi are far apart, and a uniform swing cannot see that. The baseline is a presidential race, the survey asks about House races, and it covers registered voters a month before the election. Read the map for direction and scale. It is not a forecast and not a result.
Guess first: 10 questions, 100 points
Commit to a guess before you see the map. A wrong guess made first is remembered better than a right answer read cold.
If only women voted, how many of the 50 states plus D.C. would lean Democratic under the model?
Then explore the model yourself
Women: D+14 nationally in the Pew survey. Every state is shifted +15.5 points toward Democratic candidates from its 2024 margin.
32 lean Democratic19 lean Republican7 within 3 points/ 51
- D by 10+
- by 3–10
- within 3 points
- by 3–10
- R by 10+
Changes side versus 2024 (12):
AK, AZ, FL, GA, IA, MI, NV, NC, OH, PA, TX, WI
Two things the model does show reliably. First, how lopsided some groups are: among voters under 30, the shift is large enough that even states Republicans won by 25 points in 2024 change colour. Second, how little some shifts move: with men only, the map is almost the 2024 map, because men split evenly and 2024 was close to even. The model says nothing trustworthy about any single state.
7 · Two fair readings
Two fair readings of the same data
The case that the lead was solid
- Eight points among registered voters, with an engagement advantage on every measure Pew tracks.
- Democratic-side voters were already consolidated at 82%.
- Leads among young, Hispanic and college-educated voters were wide, not marginal.
- National dissatisfaction and presidential approval pointed the same way.
The case that it was softer than it looked
- The uncommitted pool was more than twice the lead.
- The biggest undecided bloc, Republican leaners at 38%, leaned Republican by definition.
- Registered voters are not the electorate; midterm turnout skews older, where the race was close.
- A national House preference does not map directly onto 435 district contests.
Both readings use the same chart honestly. Which one better described the electorate is a question for the certified results and the post-election validated-voter studies, not for this survey.
The takeaway
Not “ahead by eight”
The chart is usually shared for its top line, and the top line was accurate as far as it went: an eight-point Democratic edge among registered voters a month out. But the structure underneath is the more durable lesson. The parties were barely trading voters. The contest in this survey was between one side that had largely come home and another whose leaners had not yet decided whether to show up for it — among registered voters a fifth of whom had still not chosen either party. That is a different, and more precise, claim than “one party is ahead by eight.”
How to read a chart like this
Six cautions
- Registered voters, not likely votersThe survey describes everyone on the rolls. The people who actually vote in a midterm are a smaller and different group.
- Subgroups carry wider errorThe full sample was 3,917 registered voters, but each row of the chart rests on far fewer interviews. A 3- or 4-point gap inside a subgroup — men, voters 65+, some college — should be treated as roughly even.
- The Asian estimate has a stated limitPew notes it represents English speakers only. It does not describe all Asian American voters.
- A snapshot, a month outFieldwork ended October 4. “If the election were held today” is not a forecast of November 3.
- National vote is not seatsThe question asks about the House race in each respondent's district, but the aggregate says nothing about how those votes are spread across districts.
- Groups overlapAge, education, race and party are tangled together. The chart shows each cut separately and cannot say which factor is doing the work.
Sources
- Pew Research Center, Midterm Vote Preferences by Race and Ethnicity, Age and Other Demographics (Oct. 8, 2026)
- Pew Research Center, With Americans in a Sour Mood, Democratic Voters Are More Engaged in Midterms (Oct. 8, 2026)
- Pew Research Center, full report and methodology (PDF)
- The American Presidency Project, UC Santa Barbara, 2024 presidential election results by state
Survey of 5,062 U.S. adults, including 3,917 registered voters, on Pew's American Trends Panel, Sept. 28 – Oct. 4, 2026. Margins, rankings and the map are calculated from Pew's published percentages; the interpretation is the author's, not Pew's. This analysis takes no position on any party or candidate.