Reading the data file…
·
Oklahoma reports each county's absentee and central ballots in a separate <county>9999 precinct bucket. For a statewide race the bucket stays with its county. County totals on this site include it, so the counties add up to the certified statewide total. For a legislative district, each candidate's bucket votes are split among the district's precincts.
Every contest on the atlas is drawn on the 1,984 current precincts. Contests from before 2026 were moved onto them by population-weighted block reaggregation. Each 2020 census block is assigned to its precinct in the old map and in the 2026 map, and votes move with the people who live in the overlap. County totals are exact. A precinct that was split is an estimate. Each candidate's central absentee votes are spread back over the county's precincts, and the 25 August runoff was handled the same way. In the Republican-only and Democratic-only contests, gold marks the candidate listed first and teal the second. Those two colours carry no party meaning. The general elections are coloured by party.
Every share on this site is a candidate's share of all votes cast in that race, other candidates included. Every margin is Democrat minus Republican in points of all votes, written with the leading party first (R+3.2). The Performance vs Governor layers take one office's margin and subtract the governor's margin on the same ballots. +D means that office's Democrat ran ahead of the Democrat for governor. +R means its Republican ran ahead of the Republican for governor.
Registration and the modelled November electorate come from the voter file. Each registrant has an isotonic-calibrated probability of voting on 3 November from the DanVAN gen26 ranker (Brier 0.0395 on the 2024 hold-out). Added up by precinct, those give expected ballots and their party mix. Turnout is measured against ALL registered voters, which is how the State Election Board counts it. The model ranks voters reliably, but its statewide total is only one estimate. The three midterms on record put it somewhere from 0.99 to 1.36 million. Shares are grouped in 10-point bands up to 100. Margins use seven bands (under 5, 5 to 10, 10 to 20, 20 to 30, 30 to 40, 40 to 50, 50 points or more), and points go in fives.
The Scenarios page is a simulator, not a forecast. It starts from the Forecast tab's numbers in every precinct and lets you change the electorate. In "Set the electorate" mode you set, for Democratic, Republican and independent registrants, the share who vote and the share of their votes that go to the Democrat. The page then recomputes every precinct: ballots are registrants times the turnout rate, and the Democrat's vote is the ballots times her share, both from the model's own precinct figures. Each precinct moves off its published forecast value by the change the levers make in the model, so at the defaults every county, district and media market equals the Forecast tab. The defaults are the forecast model's own rates and shares. The 2022 and 2018 rates in the lists are voters divided by registrants on the current roll; people who have since left the roll are missing from both counts, so those rates run high, most of all for 2018, and show the gap between parties rather than the true turnout of that year.
In "Set the margin" mode you set the statewide margin in the governor's race, and each precinct moves by its own swing slope: how far it moved per statewide point across twelve statewide Democrat-against-Republican races from 2020 to 2024 (2024 weighted 0.35, 2022 0.30, 2020 0.15), with each slope shrunk toward 1 in proportion to how little the precinct's own history says. The state is then re-centred on the margin you set. Little Dixie and the southwest carry slopes near 1.5; the metros sit near 0.9, so a statewide swing moves the rural southeast about half again as far as Oklahoma City. The presets are named starting points and each one states its source. The sensitivity chart moves one lever across its full range with the others held and shows the swing in the governor's margin. Each other office then runs off the governor's result with its own performance dial and undervote, as before.
The Early vote page counts ballots already cast, by the party each voter is registered in. Early in-person votes come from the State Election Board's in-person absentee report, read at each run the Board publishes; mail ballots come from the Board's absentee ballot totals, pulled each weekday: requested, and received back by a county election board. Every table lists all four parties and then the Democratic share of Democratic plus Republican. "Same point" means the same number of completed early-voting days: early voting runs Wednesday to Saturday before the election, so day one of 2026 (28 October) sits beside day one of 2022 (2 November) and of 2024 (30 October). A ballot cast is a ballot cast; the page never calls it a vote for a candidate, and the forecast does not read it.
The Forecast tab gives one estimate for governor statewide and one inside the 5th Congressional District, each with a range. The statewide figure starts from past governor results. It is then adjusted for the absentee ballots returned so far, and by judgment. The range comes from the spread of past governor races. Every candidate is counted. The three independent candidates together are set at 2%, below the 2014 to 2022 average for minor candidates, because there is no Libertarian candidate this year. The margin is the Republican's share minus the other main candidate's. Figures on the Forecast tab are rounded to the nearest percent, and the final figures are fixed on November 1.
The Forecast tab shows the governor's race.
The precinct map and the tables show each candidate's share of all votes cast. In every precinct the model gives the three independent candidates together their statewide share (2%) and splits the rest between the two main candidates by the precinct estimate. It does not try to show where the independents run stronger locally. The comparison on the change view is the last general for that office with the same party on the line (2022 for most offices; 2018 for Attorney General, Insurance Commissioner and Treasurer; the governor vote inside the district for the 5th District House race): that candidate's share of all votes, others included. A precinct with fewer than 50 votes in that year is left grey on the change view.
Half of the precinct map comes from a voter-level model and half from precinct history. The model is built on the voter file. It is calibrated by party and region, anchored to the 2022 regional pattern, and adjusted for public signs of support where those are known. The history is 12 statewide races from 2020 to 2024. The precincts add up to the statewide figure, so the map spreads that single estimate across the precincts. People who choose to answer the survey or send a yard-sign photo are not a random sample. The statewide figure does not come from them.
Each tab, contest and layer explains on its own page why it is there. The list below collects those notes from the data file.
The Ad buys tab combines eight public records, and each one is measured from its own documents. Ads placed (FCC orders and the two platform archives) can be added across channels, because no two of them describe the same dollar. Payments in finance reports are shown separately. They pay for those same placements, weeks later. The details below come from the tab's data file.
Incumbent party comes from the election that last decided the seat. That is usually its last general election. When no general contest followed, it is the primary or special election that settled it. A seat with no deciding election on record shows a dot in the held-by column. SD 23 and SD 41 are not up in 2026. State Senate terms are four years and staggered. The 24 even-numbered districts are up in 2026, plus SD 17 for an unexpired term. HD 92 is also an unexpired-term race, so it appears twice on the Legislature tab: once for the unexpired term and once for the next full term. A seat is deemed elected when only one candidate filed for it, so no contest appears on the November ballot. Nominees, filings, withdrawals and ballot-contest outcomes come from the Oklahoma State Election Board. Baselines are certified precinct results reaggregated onto the 2026 district lines.
On election night the Live page reads the State Election Board results feed. It scores a set of benchmark questions that were written before the count began. It does not project a winner or estimate the outstanding vote. Every figure on it is unofficial until the counties certify.
| Check | Value |
|---|
To report an error, write to the address on danthedemocrat.com.