Arizona Governor: Showing Our Work, Part Two
Arizona Governor | What we published and what happened
Showing our work, part two
On October 28 we published three figures for the Arizona governor’s race and said readers should form their own view of which to believe. Katie Hobbs won by 13.1 points. All three were wrong, and the one we built ourselves was the furthest off, by 16 points.
This post shows the gap between each figure and the result, explains what the restatement got wrong, and sets out what we are changing.
Published November 9, 2026Unofficial count of November 7Statewide canvass November 23
This is a companion to our October 28 release, which remains published and unaltered. Results are the unofficial count of November 7 and will be revised after Arizona certifies its statewide canvass on November 23.
The three figures, against the result
KStrat survey result as reported
Hobbs +5.1
Off by 8.0
American Encore turnout model, 39 / 32 / 29
Biggs +3.1
Off by 16.2
Ballots returned by October 27, 38.6 / 38.0 / 23.4
Hobbs +2.4
Off by 10.7
The actual result
Hobbs 55.60Biggs 42.48Hobbs +13.12
1,417,447 to 1,082,850. A margin of 334,597 votes on 2,549,311 cast.
KStrat as reported
H +5.1
Our turnout model
B +3.1
Ballots returned
H +2.4
Actual result
H +13.1
Biggs +5EvenHobbs +5+10+15
| What we published on October 28 | Hobbs | Biggs | Result | Error |
|---|---|---|---|---|
| KStrat survey result as reported | 48.3 | 43.1 | Hobbs +5.1 | −8.0 |
| American Encore turnout model, 39 / 32 / 29 | 44.1 | 47.2 | Biggs +3.1 | −16.2 |
| Ballots returned by October 27, 38.6 / 38.0 / 23.4 | 47.2 | 44.8 | Hobbs +2.4 | −10.7 |
| The actual result | 55.60 | 42.48 | Hobbs +13.12 | – |
Our model understated Hobbs’s share by 11.5 points and overstated Biggs’s by 4.7.
What went wrong
The arithmetic was right. The question was wrong.
Hold KStrat’s published crosstabs at a 39 / 32 / 29 electorate and you get Biggs plus 3.1. That calculation is correct, and anyone can still check it in the workbook we published with live formulas. The error was not in the computation.
We assumed the open question in the race was the composition of the electorate: how many Republicans, Democrats and unaffiliated voters would turn out. It was not. Our projection of the electorate was close. What we missed was how people voted inside their own party.
| Input to the model | What the poll measured | What the result requires |
|---|---|---|
| Biggs share of Republicans | 88.8 percent | about 78 to 81 percent |
| Hobbs share of Republicans | 2.3 percent | about 18 to 21 percent |
| Unaffiliated voters | Hobbs by 6 | Hobbs by about 28 |
about 189,000registered Republicans voted for Hobbs
about 213,000net votes lost among unaffiliated voters
334,597the final margin
Roughly one in five registered Republicans voted for Katie Hobbs, somewhere near 189,000 people. Unaffiliated voters broke against the Republican ticket by about 28 points, a net loss near 213,000 votes. No reweighting of party shares can find a Republican who votes the other way, because that voter returns a ballot exactly like a loyal one. The method we used was blind to the thing that decided the race.
Put the poll’s own two inputs into the model and it returns Biggs plus 3.1, which is what we published. Put the real ones in and it returns Hobbs plus 13.1. The model worked. It was fed two numbers that were each wrong by a wide margin, and the errors compounded.
Who got it right
Five independent public polls were released in this race. Their median was Hobbs plus 13.0 against a result of plus 13.1. The nineteen campaign and party internal polls had a median of plus 5.1.
| Poll | Fielded | Said | Error |
|---|---|---|---|
| Median of the five independent public polls | October | Hobbs +13.0 | −0.1 |
| HighGround | October 21 to 25 | Hobbs +12 | −1.1 |
| HighGround | October 6 to 8 | Hobbs +16.8 | +3.7 |
| Noble Predictive | October 26 to 28 | Hobbs +6.0 | −7.1 |
| KStrat, the poll we restated | October 25 to 27 | Hobbs +5.1 | −8.0 |
Chuck Coughlin of HighGround said publicly in mid October that 25 to 30 percent of Republicans were not backing their own party’s candidates. He was closer to right than anyone in Arizona, and his firm produced the most accurate survey taken in the state this year. We treated his polling as an outlier. It was the signal.
What we are changing
- No more restatementsWe will not publish restatements of other people’s polls at alternative turnout assumptions. The format tells a reader that the turnout model is the main uncertainty. In this race it was not, and the format itself misled.
- Ranges, not point estimatesWe will publish ranges rather than point estimates, and make them wide enough to be honest. Our working range in the final week was Hobbs plus 2 to plus 9. The result fell outside it.
- Public polls weighted at least equallyWe will weight independent public polls at least equally with campaign and party internals. In this race the five public polls beat the nineteen internals by eight points of accuracy.
- Measure defection directlyWe will measure party defection directly and repeatedly rather than inferring it from party shares. Every instrument in this race understated it, and the ballot return file cannot detect it at all.
What we will not do
We will not blame the pollster. KStrat’s published topline was Hobbs plus 5.1 and it was wrong by 8.0 points, like almost every other poll in the field. The turnout assumption, the restatement and the decision to publish were all ours.
We have left the October 28 release exactly as it was published, including the figures that turned out to be wrong. Correcting the record means adding to it, not editing it.
Method and sources
Results are the unofficial count as of November 7, 2026, from the Arizona Secretary of State. Arizona counties canvass between November 9 and 19 and the statewide canvass is November 23, so these figures will move slightly.
The estimates of Republican defection and the unaffiliated split are solved from the published result against a projected electorate of 37.9 percent Republican, 32.3 percent Democrat and 29.8 percent unaffiliated, derived from the early vote file, statewide registration of 4,374,827 and turnout of 58.3 percent. Arizona has not yet published the party registration composition of ballots actually cast, so these are estimates with a range of roughly three points on the Republican share, not certified figures. We will replace them when the state publishes the real composition.
The crossover figures are corroborated by the consortium Voter Poll of 5,023 Arizona voters, which found 11 percent of self-identified Republicans and Republican leaners voting for Hobbs and independents breaking 70 to 24 in her favor. Reconstructing the result from those crosstabs lands within six tenths of one percent of the actual count on both candidates.
