Oklahoma Governor, 2026 General Election

Results

Ballot test

Likely voters: Mazzei 46, Munson 36. Registered voters: Mazzei 38, Munson 34. Mazzei's 10-point edge among likely voters is inside the ±22-point credibility interval on the gap, so this poll cannot say with confidence who leads. 27% of registered voters are undecided. Based on 118 registered voters interviewed Sept. 27–30, 2026.

Reading these numbers

Stage 1 leaned heavily Democratic: 52% of valid respondents reported voting for Kamala Harris in 2024, compared with 20% of Oklahoma's registered voters, and only 12 reported voting for Donald Trump. Weighting corrects the balance, but each Trump voter counted about three times. Stage 2 recruited registered Republicans and 2024 Trump voters to reduce that reliance; with its 30 valid respondents, the sample holds 33 Trump voters and the design effect fell from 1.92 to 1.37. Crosstabs by party, region, age and education are not published: only registered Republicans reached 50 respondents, too few groups to report reliably.

Sensitivity checks

How much the result depends on our choices

Two checks test whether the headline depends on the Republican oversample or on how Trump voters are assumed to behave.

Check 1: stage 1 only

If stage 2's Republican and Trump-voter oversample were distorting the result, dropping it would change the numbers. It barely does; stage 2 mainly adds precision.

Check 2: Trump voters who have soured

The weights hold 2024 Trump voters at their actual share of registered voters, because the 2024 vote is a counted fact. His standing may have slipped since then, and the Trump voters who have soured on him may be less willing to take a survey. That would show up in how Trump voters vote now, not in how many of them there are. In this sample, 33 2024 Trump voters currently split 81% Mazzei, 6% Munson and 12% undecided. The table shows what happens if more of them have left Mazzei than the sample captured.

Figures are Mazzei – Munson, weighted %. Mazzei's likely-voter lead disappears if about 12% of 2024 Trump voters who back him in this sample have in fact switched to Munson. These are scenarios, not estimates; a Trump approval question in future polls will let us measure this directly.

Methodology

How this poll was conducted

This disclosure follows the items required by the American Association for Public Opinion Research (AAPOR) Code of Professional Ethics and Practices.

Sponsor

Cross Timbers Poll, a nonpartisan, independent polling project.

Conducted by

Cross Timbers Poll

Funding

This poll was paid for by Joel Katz.

Population

Adults registered to vote in Oklahoma.

Sample source

Prolific, an opt-in online research panel. This is a nonprobability sample, fielded in two stages with the same questionnaire. Stage 1 (100 completes) invited panel members whose profile lists Oklahoma as their state of residence. Because stage 1 underrepresented Republicans, stage 2 (38 completes) invited Oklahoma panel members whose profile lists Republican affiliation or a 2024 vote for Donald Trump, excluding everyone from stage 1. Both stages are combined and weighted together to statewide benchmarks, so this oversample does not inflate Republican support.

Screening

Respondents confirmed Oklahoma voter registration and gave a ZIP code. Those who were not registered, were unsure, or gave a non-Oklahoma ZIP code are excluded.

Mode and language

Self-administered online questionnaire, in English, on mobile, tablet or desktop.

Field dates

Sunday, Sept. 27, through Wednesday, Sept. 30, 2026. Stage 1 closed Sept. 27; stage 2 ran Sept. 27–30.

Incentive

Respondents received $0.35 (stage 1) or $0.50 (stage 2) from Prolific for a survey of about two minutes. Stage 2 paid more to recruit from a smaller pool.

Quality checks

Removed before analysis: failed attention check (Q9), completion time under 40% of the median, and duplicate panel IDs.

Completed responses	138
Not registered in Oklahoma or unsure (removed 19)	119
Non-Oklahoma ZIP code (removed 0)	119
Failed attention check (removed 1)	118
Duplicate panel ID (removed 0)	118
Under 37 seconds, 40% of the 1:33 median (removed 0)	118

Likely voters

Respondents who will definitely or probably vote (Q3) and who either voted in 2022 or 2024 or are following the race very or somewhat closely (Q4–Q6). 86 of 118 respondents.

Weighting

Raked to two official benchmarks. Party registration, Oklahoma State Election Board, Aug. 31, 2026: Republican 53.1%, Democrat 25.3%, Independent 20.6%, Libertarian 1.0%. 2024 presidential vote among current registrants, from certified results: Trump 42.5%, Harris 20.5%, other 1.2%, did not vote 35.8%. Respondents answering “not sure” or “prefer not to say” keep their sample share. Weights trimmed at 5.0; design effect 1.37. The sample is not weighted for age, gender, education, race or region; see Limitations.

Question order

Questions appear in the order shown below. Candidate names appear in a fixed order, which can produce a small order effect.

Limitations

What this poll cannot tell you

• It is an opt-in online sample. Respondents chose to join Prolific and to take this survey. Prolific's Oklahoma members lean Democratic: before weighting, respondents reported voting for Kamala Harris over Donald Trump in 2024 by a wide margin, the reverse of the state's result. Weighting corrects party and 2024 vote, but not every way panel members differ from other voters.
• Weighting covers two variables. The sample is weighted to party registration and 2024 presidential vote only. With about 120 respondents, adding age, gender, education, race and region would inflate the uncertainty further, so the results may be off on groups that differ within party and past vote.
• The Republican oversample came from a small pool. Stage 2 drew from roughly 114 active Oklahoma Republicans and Trump voters on Prolific. The stage-1-only check shows it did not move the result, but its respondents may still differ from Republicans generally.
• Trump voters who have soured may be missing. If they are less willing to take surveys, the poll overstates Mazzei. The defection scenarios above show how much that would matter.
• Independents lean heavily to Munson in this sample. Of 25 registered independents, 14 chose Munson and 1 chose Mazzei. That may reflect the panel rather than Oklahoma's independents.
• Likely voters are self-reported. The likely-voter group relies on respondents' own statements about voting, not on vote history from the voter file.
• Question design. Candidates appeared in a fixed order, and the lean question (Q8) was shown to everyone with a "does not apply" option, because the survey tool could not randomize or branch.

Measuring error

Why we report a credibility interval

A traditional margin of sampling error assumes every voter had a known chance of being selected. Opt-in online panels don't meet that assumption, so AAPOR advises against reporting one for them. Instead, each release reports a credibility interval: a range, based on the effective sample size after weighting, within which the true value is likely to fall if the weighting model is sound.

A gap between two candidates carries roughly twice the uncertainty of either number alone. Our releases say plainly when a lead falls inside that range.

Questionnaire

Every question, in the order asked

Respondents saw this introduction: “This is a short, anonymous survey about Oklahoma politics. It takes about 2 minutes. There are no right or wrong answers.”

Our standards

What every Cross Timbers Poll release includes

• The full questionnaire, in the order asked, including screening and attention-check items.
• Unweighted and weighted sample sizes, the design effect, and a credibility interval rather than a margin of sampling error.
• Weighting variables, target values and their public sources.
• Toplines and crosstabs for every group with at least 50 respondents.
• Counts of respondents removed at each quality check.
• De-identified respondent-level data, available to journalists and researchers on request.
• Corrections posted on the release page with the date and what changed. Past figures are never silently edited.
• No results are withheld because of what they show.

About

Who we are

Cross Timbers Poll is a nonpartisan, independent public polling project run by engaged citizens. We conduct surveys in races that might otherwise go without public polling, and we publish every result with the full methodology behind it.

We do not work for, coordinate with, or accept direction from any candidate, campaign, party or political committee. No one outside Cross Timbers Poll sees results before they are published.

Funding. Cross Timbers Poll is funded by Joel Katz. Funders have no role in choosing questions, conducting the survey, or deciding what is published.

Journalists and researchers can request the respondent-level data or ask methodology questions at:

Next poll

CTP 2026-02: pre-registered methodology

Posted Oct. 7, 2026, before data collection began. The design below was fixed in advance so that no analytic choice can be made, or appear to be made, in response to the results. Any deviation will be disclosed in the release.

Population

Adults 18+ registered to vote in Oklahoma.

Field window

Begins when the study is published on Prolific (target Oct. 7, 2026); closes end of day Monday, Oct. 12, 2026 (Central) or at 250 completes, whichever comes first. If quota cells remain unfilled at the deadline, the study stops and the achieved sample is reported with the shortfall disclosed. No mid-field changes to quotas, targets or questions.

Sample source

Prolific opt-in online panel, Oklahoma residents, with everyone who took CTP 2026-01 excluded. A single quota sample set at launch on Prolific's self-reported U.S. political affiliation: Democrat 62, Republican 48, Independent 95, Other 10, None 25, Prefer not to say 10 (n = 250). The Republican quota is capped by the available pool (79 active members); party registration is corrected in weighting. Prolific offers no 2024-vote attribute, so the 2024 vote is handled entirely by the raking targets below.

Instrument

Google Forms, linked from Prolific. Candidate order on the ballot question is randomized per respondent. The candidate-lean follow-up is shown only to respondents who answer Undecided. Prolific IDs are captured by URL for payment and deduplication only; no names or emails are collected.

Compensation

$0.70 per complete (about $21/hour at the 2-minute estimate), Prolific's competitive pay band for this audience.

Changes from CTP 2026-01

Randomized candidate order; lean question branched to undecideds only; two benchmark items added — Trump job approval (measures Trump-voter defection directly instead of modeling it) and the 2022 governor vote (Stitt 55.5%, Hofmeister 41.8% certified), both reported against known or independently measured values regardless of how they come out. All other questions identical for trend comparability.

Exclusions

Applied in order, with counts reported: not registered in Oklahoma or unsure; ZIP outside 73001–74966; failed attention check; duplicate Prolific ID; completion time under 40% of the median. Returned, timed-out and rejected submissions are excluded automatically.

Weighting

Raking on two margins among valid registered respondents, weights trimmed at 5.0: party registration (OK State Election Board, Aug. 31, 2026: Republican 53.1%, Democrat 25.3%, Independent + Libertarian 21.6%) and the 2024 presidential vote as a share of current registrants (certified: Trump 42.5%, Harris 20.5%, other 1.2%, did not vote 35.8%). "Not sure" and "prefer not to say" keep their sample shares. If the valid sample reaches n ≥ 200, age (18–29/30–44/45–64/65+) and education (4 bands) are added as third and fourth margins using ACS 1-year estimates for Oklahoma adult citizens; below 200 they are reported unweighted as a composition check only.

Likely voters

Will definitely or probably vote, AND voted in 2022 or 2024 or follows the race very or somewhat closely. Registered-voter and likely-voter results are both reported.

Reporting

Topline with effective sample size, design effect and 95% credibility intervals; leaned ballot; Stitt and Trump approval; ballot by Trump approval and by 2024 vote; both benchmark checks; full question wording; and this pre-registration, unchanged.

All polls

Release archive