Methods
Sample design
Respondents were drawn from the L2 Oklahoma voter file and invited by text message, sent through RumbleUp, to a web questionnaire. Three samples were drawn:
- Statewide: 40,000 records drawn Oct. 24 from a universe of 422,839 active registrants with a mobile number, one per household, across all levels of vote frequency.
- 5th District supplement: 10,000 records drawn Oct. 27 from the district's registrants not already drawn statewide.
- Runoff-voter reserve: 10,000 records drawn Oct. 27 from voters the file shows voting in the Aug. 25, 2026 runoff, not already drawn in either other sample. The reserve was released under a pre-registered rule triggered when fewer than 150 self-reported Drummond runoff voters had responded by the end of the second full day of fielding.
Statewide toplines use the statewide sample alone. Pooled tables combine all three samples using selection-probability base weights with a multiplicity correction: each record's base weight is the reciprocal of its probability of inclusion in any of the three samples, so a voter eligible for more than one draw is not counted twice. The pooled likely-voter governor result (n=976) is Mazzei 46, Munson 43. The 5th District table is raked to the district's party registration and county totals.
Methods
Inclusion and likely voters
A respondent is included if they confirmed their registration, answered the governor question, matched the voter file, and started before the 11:59 p.m. Oct. 28 close. Pre-registered exclusions remove full completions under 90 seconds and straightliners (below). One survey started after close was excluded. Of 1,019 included respondents, 943 completed the survey and 76 are partial; 596 are statewide.
A respondent is a likely voter if they (a) report having already voted; (b) say they will definitely or probably vote and the file shows a vote in the 2022 or 2024 general election; or (c) say they will definitely vote and registered after Nov. 5, 2024. Statewide, 575 of 596 respondents are likely voters.
Methods
Weighting
Weights are computed by iterative proportional fitting (raking) on the registered-voter base. After each full pass, weights are trimmed at 5.0 times the mean, re-centered and raked again, until every marginal matches its target to within 0.0001 of a share. Categories without a target (gender not recorded on the file; respondents who preferred not to report a 2024 vote) keep their sample share. Likely-voter tables use the registered-voter weights restricted to likely voters and renormalized.
The estimate is raked to party registration (Oklahoma State Election Board, Sept. 30, 2026), age, gender and region (L2 counts of all active registrants, received Oct. 28 before close), and recalled 2024 presidential vote. The recall target is the certified 2024 result (Trump 66.2, Harris 31.9, other 1.9), applied only to respondents whose file record shows a 2024 general-election vote and who reported one. It converges in 24 iterations, with 10 weights at the trim.
The alternative uses the same targets without recalled vote. It converges in 6 iterations, with none at the trim.
Respondents aged 18–34 are 3% of the statewide sample against a 23.6% target, too few to rake within the trim. For raking, 18–34 and 35–49 are combined into one 18–49 band. The four-band age distribution is published unweighted.
The pre-registration specified recalled vote as a share of all registrants, including the 35.8% who did not vote in 2024. Because the sample is drawn from active registrants with a mobile number, 91% of whom voted in 2024, 2024 non-voters are 5% of statewide respondents by the voter file; the pre-registered target (35.8%) describes all registrants, including inactive ones, and that rake does not converge. It is not published as an estimate; it is recorded in the log.
Targets and achieved marginals
On the registered-voter base, both weights match every raking target below. Party on the likely-voter base:
| Party registration | Target | Unweighted LV | Estimate LV | Alternative LV |
|---|---|---|---|---|
| Republican | 53.3 | 50.8 | 53.3 | 53.1 |
| Democrat | 25.4 | 35.1 | 25.9 | 25.9 |
| Independent | 20.4 | 13.6 | 20.2 | 20.3 |
| Libertarian | 1.0 | 0.5 | 0.6 | 0.7 |
| Other raking targets | Target |
|---|---|
| Age 18–49 | 46.8 |
| Age 50–64 | 23.5 |
| Age 65+ | 29.7 |
| Female | 52.8 |
| Male | 47.2 |
| Oklahoma City metro | 36.3 |
| Tulsa metro | 26.1 |
| Rest of state | 37.6 |
The OKC metro is Oklahoma, Cleveland, Canadian, Logan, McClain, Grady and Lincoln counties; the Tulsa metro is Tulsa, Rogers, Wagoner, Creek, Osage, Okmulgee and Pawnee.
Recalled 2024 vote
| Likely voters who reported a 2024 vote | Trump | Harris | Other |
|---|---|---|---|
| Unweighted | 46.7 | 49.7 | 3.6 |
| Alternative weight | 47.1 | 46.8 | 6.1 |
| Estimate | 65.0 | 33.1 | 1.9 |
| Actual 2024 result | 66.2 | 31.9 | 1.9 |
Respondents reported a 2024 vote far closer than the state's result, and party registration alone does not correct it. That is the difference between the two weights.
Methods
Margin of error
Margins of error are 1.96 × √(deff × 0.25 / n), using the Kish design effect of each weight. For statewide likely voters (n=575): estimate deff 1.95, effective n 295, ±5.7; alternative deff 1.61, effective n 357, ±5.2. Subgroup margins are larger and are shown in each table. A bootstrap 95% interval for the likely-voter governor margin (Munson minus Mazzei) runs from −13.1 to +8.1 under the estimate and from +8.2 to +27.3 under the alternative.
Methods
Choosing the estimate
Public pollsters in 2026 divide on this (methodology statements reviewed Oct. 29: Emerson, co/efficient, Recon/Siena, YouGov, NYT/Siena, Pew). Emerson, co/efficient and Recon/Siena weight state polls to party registration without recalled vote — the approach of our alternative. YouGov and the New York Times/Siena weight recalled vote to the 2024 result among confirmed voters — the approach of our estimate.
The pre-registration named the recall weight as primary. During review we considered reversing that. We kept it on a down-ballot test. Under the alternative, four of the five Republicans on the statewide ballot fall by 9 to 11 points (the Superintendent's race by 5): the Senate race is tied (Hern 39, Thomas 39), the Attorney General's race leans Democratic (Coffey 51, Echols 40), and the Corporation Commission race is even (Boles 44, Eastman 43). We read a shift of that size and uniformity across contests as a sample leaning Democratic within party, rather than as five races moving together; that is a judgment, and the alternative weight is published so readers can make their own. Under the estimate, the down-ballot races return to customary Republican margins while the governor's race does not.
The down-ballot comparison is a diagnostic, not a separate weighting. The same weight applies to every respondent on every question; the governor's race is not weighted differently from any other. We read a parallel shift across all five statewide races as a sample that leans Democratic within party, because the respondents' own recalled 2024 vote (Harris 50, Trump 47, unweighted) leans the same way in a state that voted Trump 66–32, and coattails do not change what people did two years ago. A reader who instead reads the shift as a Democratic wave reaching the down-ballot races should prefer the alternative weight, which is published for that reason. Under either weight, the implied ticket-splitting is modest: about 3 percent of likely voters choose the Republican for Senate but not for governor.
Methods
Straightlining sensitivity
The pre-registered rule removes respondents who gave identical positions on every approval and support scale and finished in under three minutes. It removed 65 respondents, 39 statewide (19 Democrats, 14 Republicans, 6 independents); 59 of the 65 gave strongly negative answers throughout. The rule is applied as registered. Restoring the 39 and re-weighting, with the recalled-vote correction applied to them as to every other respondent, moves the registered-voter governor margin less than one point toward Munson under the estimate and about three points under the alternative (Munson 56, Mazzei 34). Corrected Oct. 30: this note first gave the estimate as Munson 46.6, Mazzei 41.9, a calculation that left the restored respondents' recalled vote uncorrected. Most of those removed gave the same strongly negative answer on every scale, and 58 of the 65 chose Munson for governor, which suggests the rule caught many genuine respondents; it is applied as registered and will be revised for the next survey.
Methods
Examined and not used
Raking Aug. 25 voters to the certified Republican runoff result (Mazzei 50.27, Drummond 49.73) was examined, because voters the file shows voting on Aug. 25 are 62% of statewide respondents. It requires cutting that cell by two-thirds and raises the design effect to 2.5, and the file’s Aug. 25 flag also counts votes on the two state questions and in Democratic runoffs, so it cannot be matched to the Republican result. It is not used.
Methods
Comparison with earlier CTP polls
CTP 2026-01 and 2026-02 were fielded in September and October on Prolific opt-in panels and found Mazzei ahead by 10 and 3 points among likely voters. This survey uses a voter-file frame and text-to-web mode, so the three are not a trend line. The estimate is consistent with them; the alternative is not.
Methods
Limitations
- Response rate: 2.0% of delivered invitations (1.8% completes only). No incentive was offered.
- Aug. 25 voters are over-represented: the voter file shows 36% of the statewide frame voting on Aug. 25, when the ballot carried party runoffs and two state questions open to all voters, and those voters responded at a higher rate, so 62% of statewide respondents did. The Drummond-voter table is shown under both weights, for registered Republicans only.
- Age: 3% of statewide respondents are under 35; the youngest voters are weighted as part of an 18–49 band.
- 5th District: the district's recall target (Trump 58.0, Harris 39.8) is the certified 2024 result by district; no district age or gender targets were available, so it is weighted on party and county, with recalled vote corrected among respondents the file shows voting in 2024.
- Voter-file demographics: party, age, gender and region come from the voter-file record of the voter who was texted, not from the respondent. If someone else in the household answered, those fields describe the wrong person. Respondents confirmed they were registered; fewer than 1% of those the file shows voting in 2024 say they did not vote, and among registered Republicans, 8 of the 204 the file shows voting in the Aug. 25 runoff say they did not. The survey did not ask age or gender to check the record directly. Separately, 61 respondents registered as Democrats or independents said they voted in the closed Republican runoff, which they could not have done (Day 2 correction).
- Regional estimates: recalled 2024 vote is corrected statewide, not by region. Against certified 2024 results by county, the estimate's weighted recall runs low for Trump in the Oklahoma City metro and slightly high in the rest of the state (Day 2).
- Recall as an anchor: weighting to a vote cast two years earlier can over-correct toward the Republican candidate when opinion has shifted since; this is a recognized concern among pollsters this cycle. The alternative is published so readers can see the size of that adjustment.
Methods
Pre-registration and disclosure
The design, questionnaire, inclusion rules, likely-voter definition and weighting targets were pre-registered Oct. 24, 2026. Every decision made after close — the age band, the recall implementation, the choice of estimate, the multiplicity correction and the runoff rake — is in the public log with its time and reason.
We use an AI assistant (Anthropic's Claude) for drafting, correspondence and writing and reviewing analysis code; sampling, targets and rules were fixed in the pre-registration; every published number is checked against the data before release. No synthetic respondents, no imputed answers.
About
Who we are
Funding. Cross Timbers Poll LLC is registered in Oklahoma and is funded by members who are private citizens with no campaign, party or committee role. It does no work for candidates, campaigns, parties or committees.
Questions about these methods: contact@crosstimberspoll.org. Results: CTP 2026-03 Day 1. Pre-registration: CTP 2026-03 pre-registration.