PROJECTED PARTICIPANTS
—of registered voters
Historical-error stress range
Illustrative range · no calibrated probability coverageNOVEMBER 3, 2026 · OKLAHOMA COUNTY
The turnout outlook
Who may participate. What the evidence can tell us.
HOW TO READ THIS FORECAST
A historical baseline,
with its limits in view.
Current registrations × a blend of 2018 and 2022 participation rates. The 70/30 weights are inherited assumptions, not a proven optimum.
Examine the evidence ↗Explore a result scenario →The participating electorate
Registered party ≠ candidate choiceSINCE THE PREVIOUS RELEASE
A new archive begins
View every saved forecast →ABSENTEE OBSERVATIONS
requests on file
recorded receipts
Receipts are not counted votes. These observations do not mechanically raise the forecast.
Inspect ballot status →Election participation can exceed votes cast in any one contest. This model does not forecast candidate preferences or winning probabilities.
CONDITIONAL RESULTS · YOUR ASSUMPTIONS
Scenario lab
Translate an electorate into vote totals under explicit assumptions.
Oklahoma County Assessor: Candidate A = Larry Stein (Republican); Candidate B = John Gibbons (Democratic). Names verified against the official November 2026 election list on September 25, 2026. Starting independent shares are 55% Democratic / 45% Republican by user request; other registration groups start at an illustrative 50/50. These are scenario assumptions, not measured Assessor preferences. Changes here never alter the published turnout forecast.
Independent vote-share assumption
USER ASSUMPTION · NOT CALIBRATEDThe starting split gives John Gibbons (D) 55% and Larry Stein (R) 45% of valid Assessor votes from registered independents. This changes candidate preference assumptions, not turnout.
Evidence, sample sizes and limitations
The GBAO CD5 memo, released by the Nelson campaign, reports 55% Nelson support among independents in a September 17–20 survey of 500 likely voters. It does not establish registered-independent status or the subgroup sample size. The Republican 45% is our assumed remainder, not a poll finding. Different geography and candidates limit its transfer to this race; the full-sample margin of error does not apply to this subgroup.
The 2018 CCES contains 60,000 national respondents, 165 in Oklahoma County, and 23 active registered independents matched to Oklahoma records. Only 16 of those have post-election weights and validated turnout. This small subset does not establish an Assessor preference estimate. Historical replays retain their original countywide results.
Set the assumptions
0
The remaining participants voted in the election but have no valid Assessor vote in this scenario. This can include skipped contests, invalid votes or ballot eligibility differences; aggregate totals do not identify the reason. This is separate from registered voters who did not participate.
Uses the historical election selected in the benchmarks below. Starting contest participation is 100%; lower it to account for contest nonvotes.
| Registration | A · Larry Stein (R) % | B · John Gibbons (D) % | Other % |
|---|
RESULT UNDER THESE ASSUMPTIONS
A · Larry Stein (R)
B · John Gibbons (D)
Other valid votes (hypothetical)
Participated, no valid Assessor vote
| Outcome | People / votes | % of participants | % of valid Assessor votes |
|---|
Turnout sensitivity
The same vote shares, applied at the low, central and high turnout stress settings.
| Participants | Valid Assessor votes | No valid Assessor vote | Stein − Gibbons |
|---|
Independent-share break-even
Holds all other groups and the independent “Other” share fixed. A gain for A comes from B. This is arithmetic, not campaign advice or a forecast.
Independent-share sensitivity
| A · Larry Stein (R) share | B · John Gibbons (D) share | Stein − Gibbons votes |
|---|
OKLAHOMA COUNTY ASSESSOR · OBSERVED RESULTS
Historical benchmarks
Replay a past countywide vote split and its share of participants casting a valid Assessor vote. This holds historical behavior fixed; it is not a prediction of the 2026 candidates. No party loyalty or crossover is inferred.
| Election | Republican | Democratic | Valid votes / participants | Participated, no valid Assessor vote | Source |
|---|
Loading historical results…
Uses the election-participants slider above and the selected year's observed contest participation rate. Manual vote-share and contest sliders apply only to the separate scenario above.
286 POPULATED PRECINCTS · ALL-PARTY PARTICIPATION
Precinct explorer
Explore the geography behind the county estimate.
SELECT A PRECINCT
Every area,
one county.
Click an area or a row below to inspect its turnout and absentee observations.
Use the search field to isolate a precinct. Arrow keys pan the focused map; + and − zoom.
Precinct patterns come from the released statewide model and are adjusted to match this county total. Stress sensitivity reflects shared county uncertainty, not a validated precinct confidence interval. Boundaries were supplied and confirmed unchanged since 2022.
| Precinct | Registered | Participants | Turnout | Requests | Receipts |
|---|
OBSERVED BALLOT STATUS
Absentee & early voting
Keep requests, receipts and participation distinct.
A missing receipt can mean a ballot is in transit or awaiting an update. A recorded receipt does not establish acceptance. Requests are not votes and cannot simply be added to the turnout estimate.
Early participation
Official absentee & early-vote reports ↗Signal quality
Final historical request files are not comparable to a September snapshot. A same-days-before-election series is needed before fitting a predictive adjustment.
Source checks
The source checker records availability and flags new reports. Report availability alone does not verify a count; numerical updates require a validated aggregate import.
Loading source status…
VERSIONED PUBLIC RELEASES
Forecast archive
Track changes without rewriting the past.
| Published | Source date | Participants | Change | Model | Snapshot |
|---|
Before this archive
The earlier published 70/30 test projected 252,867 participants. Its original data is preserved separately; it is not presented as a newly reconstructed historical time series.
Download the original published test ↓HISTORICAL EVIDENCE · EXPLICIT LIMITS
Backtesting
What happened when an earlier participation pattern met a later election?
These hindcasts use later-election registration as an input, but never use later participation to generate the prediction. They test historical rate transfer conditional on known registration, not a fully reconstructed advance forecast. The 70/30 blend itself has no untouched later midterm to test.
| Train → evaluate | Predicted | Observed | Error | Rate error | Uniform-rate baseline error |
|---|
Where the range comes from
Party and precinct stress bounds are reconciled to the same county bounds. They move together; treating them as independent intervals would understate shared election uncertainty.
What would establish calibration?
Additional historical midterms, dated pre-election registrations, and forward-held-out evaluations. Coverage must be scored across elections, not thousands of correlated precincts treated as independent elections.
The model does not publish a nominal 80% or 95% interval with the current evidence.
Observed participation history
| Election | Registered | Participants | Turnout | Republican | Democratic | Independent | Libertarian |
|---|
Historical-weight sensitivity
Apply alternative 2018/2022 weights to the same current registrations. These are alternative assumptions, not uncertainty bounds.
| 2018 / 2022 | Participants | Turnout |
|---|
Earlier party-composition checks
The original site also tested party allocation after supplying the actual county turnout total. These smaller errors measure composition only and do not validate an advance county forecast.
| Check | Weighted rate error |
|---|
The original Libertarian errors were 15.45 points in 2022 and 6.79 points in 2024; county averages can hide small-group errors.
Party-level hindcast errors
| Election | Registration | Predicted | Observed | Error |
|---|
REPRODUCIBLE INPUTS · NEUTRAL ASSUMPTIONS
Sources & method
Understand the model before using its numbers.
Aggregate sources→Validated snapshots→Turnout model→Public forecast
Turnout forecast
For each registered party: current registrations × (0.70 × 2018 turnout rate + 0.30 × 2022 turnout rate). The sum is county participation. Current active and inactive registrations are included and held fixed through Election Day.
The weight choice preserves the previous published test. It has not been independently validated or selected for superior predictive performance.
Precinct estimates use a common shift in historical model turnout odds to reconcile the precinct sum to the county total. No party preference is inferred within a precinct.
Conditional result scenarios
A means Larry Stein (Republican) and B means John Gibbons (Democratic) for Oklahoma County Assessor. Users specify election participation, valid contest participation, and A/B vote shares by registration group. The unallocated candidate share goes to hypothetical “Other” valid votes. Participants without a valid Assessor vote are calculated separately using the contest-participation control; they are never counted as “Other” votes. Registered nonparticipants are a separate remainder. All result totals derive only from those assumptions.
A common odds adjustment distributes user-specified turnout across registered parties while respecting registration limits. There is no polling model or inferred candidate preference.
Historical replays use official 2018 or 2022 countywide Assessor results and observed contest participation. They do not estimate registration-group vote preferences. Two elections do not establish a calibrated candidate forecast. Scenarios stay in the browser. Saving stores them on this device; exporting produces a portable record with the forecast version.
Sources and freshness
| Input | Source | As of / availability | Role |
|---|
Uncertainty & limitations
The displayed range uses one prior midterm's county rate error. It is not a calibrated probability interval. Forecast errors can be larger, and registration can change after the source date.
Current voter history excludes people removed from the current register. It is not substituted for the original historical electorate. No individual propensity scores or contact lists are published.
Absentee requests, receipts and early-vote observations remain separate from model estimates until predictive adjustments can be validated.
Release integrity
Inputs must reconcile across party, precinct and county totals. Invalid imports are rejected. Published snapshots retain their input and model versions. Older source dates cannot silently replace newer observations.
Source dates describe the evidence; publication dates describe model releases. A fresh check does not make an old source current.