ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
Prolific Poll (US), October 2026: full results
2
Accompanies our article on the October 2026 generic congressional ballot and the choices behind our estimate. Every estimate reported in the article and its charts is in sheets A1–A8, with its margin of error and unweighted base; the numbered sheets give weighted toplines and crosstabs, by party identification and 2024 presidential vote, for every political question asked. The full questionnaire is published alongside this workbook.
3
Survey
4
Prolific surveyed 1,518 US adult citizens (analysed) on 21–22 October 2026, online, through Prolific's opt-in participant panel. Participants were invited using propensity-score sampling: a statistical model of how Prolific's US panel differs from the US adult population (2024 American Community Survey) on age, sex, race, education, region and 2024 and 2020 presidential vote defined 25 groups, each invited in proportion to its population share. Design weights correct for groups that came in above or below plan. 2020 presidential vote comes from Prolific's participant profile data.
5
Data quality: of 1,523 completed interviews, 1,518 are analysed; 4 respondents who said they were not US citizens and 1 duplicate session were excluded. 1,517 of the 1,518 passed an attention check, and the median completion time was about 3.5 minutes.
6
Weighting
7
Estimates are weighted by raking (iterative proportional fitting) to: age × sex, race × education, region and employment (2024 American Community Survey, adult citizens); household income in four bands (Pew Research Center NPORS 2026); 2024 presidential vote (certified results); 2020 presidential vote; registration (Census Bureau Current Population Survey, November 2024: 73.6% of adult citizens); and party identification (Pew NPORS). Weights are trimmed at 5 times the mean. The preferred weights use the 2026 NPORS party identification shares, counting independents who lean towards a party with that party; sheets A3–A4 also show the 3-year (2024–2026) average and the versions without leaners.
8
Populations
9
Registered voters: respondents who say they are registered to vote at their current address. Likely voters (preferred): every respondent is weighted by their modelled probability of voting, from a model of validated turnout in the 2014, 2018 and 2022 Cooperative Election Study that uses past vote, registration, age, education, race, sex and stated intention to vote; turnout by age and education is matched to the Census Bureau's 2022 pattern corrected to actual state turnout (Hur and Achen), and overall turnout is fixed at 47% of adult citizens. Other likely-voter definitions are in sheet A4.
10
Counting answers
11
Generic ballot: respondents who answered "Not sure" were asked which party's candidate they lean toward and are counted with that party. "Other / not sure / won't vote" combines the remaining answers. D − R is the Democratic share minus the Republican share, in percentage points.
12
Margins of error
13
Our sample is not a probability sample and is weighted, so the textbook margin of error does not apply. Following Sturgis and Kuha (Journal of the Royal Statistical Society, 2017), we estimated sampling variability by bootstrapping: respondents were resampled with replacement within the 25 sampling groups 500 times, and the whole estimation (all weighting and the turnout model) was repeated on each resample. ± is 1.96 times the bootstrap standard error (95% confidence); "95% CI" is the 2.5th to 97.5th percentile of the bootstrap estimates. The experiment table uses 2,000 resamples within each column, unweighted. These margins cover sampling and weighting variability only: not the choice of party targets, turnout level or method (shown as alternatives in sheets A3–A8), and not other sources of error every poll faces.
14
Bases
15
Unweighted n: number of respondents in the base. For likely voters, every respondent contributes in proportion to their probability of voting, so the unweighted n is all respondents with a non-zero probability. Weighted n: sum of the weights in the base (weights average 1 across all 1,518 respondents; for likely voters, weight × probability of voting). Effective n: the sample size a simple random sample would need for the same precision, given the variation in weights. Columns with fewer than 50 respondents are shown in grey italics.
16
Subgroups: estimates for subgroups rest on smaller effective samples and carry wider margins of error; differences between subgroups are often not statistically distinguishable. Our weights correct each weighting variable overall, not its relationship with vote choice within every subgroup.
17
Why the likely-voter n is larger than the registered-voter n: the turnout model gives every respondent a probability of voting, including the small number who say they are not registered (registration is self-reported, and many states allow same-day registration). Their probabilities are low, so they carry little weight, but they count in the unweighted n.
18
Why our margins of error are smaller than a design-effect formula would give: formulas such as 1.96 × √((1 + CV²) / n) count only the precision lost through unequal weights. Weighting to variables that strongly predict the answers (party ID, past vote) also removes variability, and the bootstrap captures both effects.
19
Rounding
20
All numbers are shown to one decimal place (the article rounds for readability). Shares may not sum to 100 because of rounding. A figure of 0.0 is usually a rounding artefact.
21
Sources
22
All estimates are computed in R. Wording of every question is taken from the fielded questionnaire.
23
Contact: oriol.jover@prolific.com
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100