The Same Pollster Had the Republican Up 7 in 2022, and He Lost. Here's Every 2026 Poll of the Attorney General Race, Then the Count.

On October 31, 2022, SurveyUSA put out its last poll of the Minnesota Attorney General race for KSTP. Jim Schultz 49, Keith Ellison 42. Eight days later Ellison won by 20,815 votes.

On October 29, 2026, the same pollster, working for the same station, put out its last poll of this race. Ellison 49, Ron Schutz 41.

So which one do you believe? Neither, yet. Tonight Minnesota counts, and the count is the only poll with a sample size of everybody who voted. It's also the only one nobody has to weight.

One plain disclosure before the facts. I ran against Keith Ellison in the August primary and lost. I've also said publicly that I'm supporting him in this one. Weigh what follows with that in mind. Everything below comes from records you can open yourself, and they're listed at the bottom.

Every public poll I could open

This is every general-election poll of Ellison against Schutz I could read in the pollster's own release, or, for one of them, the sponsor's. "LV" is likely voters. "RV" is registered voters, a bigger and looser crowd.

The June Ellison and Schutz numbers come from SurveyUSA's own July release, which reports that "Last month, incumbent DFL candidate Keith Ellison led attorney Ron Schutz by 7 points, 44% to 37%". The June undecided share is from KSTP's June story. State Navigate's 52–44 counts leaners. Those are the folks who said they were undecided and then, pressed a little, admitted which way they tilt. So it isn't built like the others, and its text doesn't print an undecided share.

Don't read these as three takes on one poll. SurveyUSA surveys people online from a commercial panel, Cint USA, and weights the results. State Navigate texts voters pulled from the L2 voter file and sends them to a SurveyMonkey questionnaire. Morning Consult's sponsor reports registered voters, not likely ones. Three pollsters. Three different crowds.

Now look at the column SurveyUSA calls error. It won't call it a margin of error. It calls it a "credibility interval," and its own fine print says why: it "better reflects the sampling uncertainties associated with gathering some percentage of respondent answers using non-probability sample." The same fine print admits that sampling error "is only one of many types of error", and that "Non-sampling errors cannot be quantified."

I had to read that last line twice. That's a pollster telling you, in writing, to hold its numbers loosely. I'd take it up on the offer.

And the interval sits on each candidate's number, not on the gap. State Navigate's release says it outright: its ±3.5 sits "around each candidate's result." If one candidate is 4.2 points high and the other 4.2 low, SurveyUSA's 8-point lead could really be a tie. That's grade-school arithmetic. Good luck fitting it in a headline.

Ten points in six weeks

In mid-September SurveyUSA had Schutz ahead, 45 to 43. Its September release said Ellison "has consistently polled between 42% and 44%", and that Schutz had climbed out of the 36-to-38 range "to tie Ellison at 43% last month and now to the lead at 45%."

Six weeks later the same pollster had Ellison up 8. That's a 10-point swing on the margin, in a race where, by SurveyUSA's own count, Schutz "has never polled above 45%". By mid-October SurveyUSA itself was calling the contest "very close and recently quite bouncy". Bouncy. I can't improve on that.

So what moved? Skip the headline and go to the bottom of each table, where SurveyUSA prints who it actually surveyed.

KSTP's own story caught that last row: "That +8 advantage for Democrats is double what it was two weeks ago." Credit to the station for printing it. A topline fits in a headline. The sample usually doesn't.

Why does the party mix bounce around like that? In September, at least, SurveyUSA said it weighted its respondents "to US Census ACS targets for gender, age, race, education, and home ownership." Party isn't on that list. Its standing methods statement says it "may or may not weight to Party ID, depending on client preference." Read that one again. Whether the poll gets weighted to party is the client's call. So the share of Democrats and Republicans can float with whoever answered that week.

How much of the swing does that explain? A lot. Among Republicans and Democrats, the final poll barely budged from September. Republicans went 91 percent for Schutz in September, 91 in mid-October and 91 at the end. Democrats went 87, 89 and 87 for Ellison. The partisans sat still. Take the final poll's answers inside each party group and pour them back into September's party mix, and Ellison's lead drops from about 8 points to about 4. By my arithmetic, roughly four of the ten points came from who was in the sample. And the last time the sample ran D+8, in early August, Ellison was up 8 then, too.

Nearly all the rest came from independents. In September they favored Schutz 44 to 34. At the end they favored Ellison 47 to 35, with 18 percent still undecided. That's a 22-point turn on the margin. It's also about 160 people, if you multiply the independents' 23 percent share by 714 likely voters. That's a lot of freight for a group that size to carry. Whether independents changed their minds, or a different set of independents answered the survey, the release can't tell you.

Women moved too. SurveyUSA's final release has Ellison's lead among women at "18 points, up from 12 points two weeks ago and 5 points in September", and Schutz's lead among men down from 10 points to 2.

Who's in the numbers

The final SurveyUSA poll splits the race the way you'd guess, just more so. Ellison led 53 to 35 among women; Schutz led 47 to 45 among men. Voters 18 to 34 went 60 to 27 for Ellison. Voters 65 and older went 47 to 45 for Schutz. College graduates went 61 to 30 for Ellison, voters with a high-school education 52 to 36 for Schutz.

Geography is the whole state in miniature. The Twin Cities region, 60 percent of the sample, went Ellison 57 to 33. Western Minnesota went Schutz 65 to 24. Northeastern Minnesota went Schutz 50 to 39. Southern Minnesota, Ellison 51 to 43. In SurveyUSA's urban, suburban and rural columns, urban voters went Ellison 61 to 28, suburban voters Ellison 54 to 38, rural voters Schutz 58 to 31.

That suburban number is the one I'd watch tonight. In September's SurveyUSA poll the suburbs were Ellison 45, Schutz 43. Six weeks later it's a 16-point lead. That's a big move for one slice of one survey. SurveyUSA's release doesn't say how it sorts suburban voters, and its suburbs aren't a county line, so they won't line up neatly with the count. The county count is coming anyway, and it doesn't care how anybody sorts anything.

Morning Consult, polling registered voters three weeks earlier, drew the same map with more fog on it: suburbs Ellison 45 to 41, rural Schutz 43 to 30, urban Ellison 58 to 29. Its sponsor reported 14 percent undecided overall, "including 22% of voters ages 35–44 and 17% of women", and said the race had "more undecided voters than either the governor or Senate contest."

State Navigate went looking for ticket-splitters. Its release says Ellison runs about 2 points behind Amy Klobuchar in the governor's race, and that "4% of Klobuchar voters back Schutz, while only 1% of Demuth voters back Ellison. Much of it is in Greater Minnesota, where Ellison trails by 15 against Klobuchar's 11."

One more thing, and this one made me laugh. SurveyUSA and State Navigate were both in the field October 26 through 28. SurveyUSA's release says 20 percent of its likely voters had already voted. State Navigate weighted to 15 percent. Two polls, in the field together for three days, and they couldn't agree on how many ballots were already in the box. That doesn't prove anybody screwed up. Different crowds look different.

2022 is the yardstick

Here's the test I'd put every 2026 number through. Same pollster, same sponsor, nearly the same question, four years ago.

The State Canvassing Board's certificate has Ellison at 1,254,371 votes and Schultz at 1,233,556, with 2,374 write-ins. A 0.84-point win. The final poll had Schultz up 7. It missed the margin by 7.84 points. That's about twice the ±3.9 the poll put on each candidate's number, the very edge of what that interval allows on the lead. The early-September poll missed the other way, by about 5. Wrong in both directions in one fall. That takes some doing.

Now look at where the miss was. The poll had Schultz at 49. He got 49.53 percent. It had Ellison at 42. He got 50.37. Ellison's number was eight points short and Schultz's was half a point short. The poll's numbers include the 9 percent who were undecided and the count doesn't, so for the two to line up, nearly all of the undecided had to end up with Ellison, or the sample held too few of his voters. Either way, the whole miss was on the incumbent's side.

This part I didn't expect. The same 2022 poll had Tim Walz beating Scott Jensen "by 8 points, 51% to 43%." The certificate has Walz winning by 192,408 votes, 7.91 points of the two-candidate vote. Same respondents. Same week. The pollster nailed the governor's race and blew the attorney general's race by almost eight points. That 2022 poll had Ellison running 15 points behind Walz. On the actual ballots he ran about 7 behind. The gap between the two races was real. The poll just showed it about twice as wide as the ballots did.

And the 2022 sample leaned the other way. SurveyUSA's final 2022 poll was 36 percent Republican and 33 percent Democratic. Its final 2026 poll is 41 percent Democratic and 33 percent Republican. The party mix moved 11 points between the two finals. Maybe Minnesota changed that much in four years. Maybe the panel did. One poll can't tell you which, and neither can I.

So I'm not telling you the 2026 polls are wrong. I'm telling you this pollster missed this race by about 8 points the last time, with a smaller stated error than it carries now, and that a down-ballot race with about 10 percent undecided is exactly where it missed.

How close is close

The 2022 race was decided by 20,815 votes. That's 0.84 points. That's a rounding error with a certificate attached.

The only closer attorney general race I found in the official canvass records is 1956. The Legislative Manual for that year has Miles Lord, marked elected, at 694,694 votes and Keith Kennedy at 688,931. That's 5,763 votes, about 0.42 points of the two-man vote.

The other years I pulled to the official total weren't nearly as close. Lord won again in 1958, 627,851 to 498,478, about 11.5 points. Lori Swanson won in 2014 with 52.60 percent to Scott Newman's 39.01. Ellison's first win in 2018 was 1,249,407 to Doug Wardlow's 1,150,459, a 98,948-vote margin, 3.89 points of the vote for the three candidates on the ballot.

You'll hear 2022 called the closest since 1956. It may be. The other twelve elections from 1962 through 2010 are sitting in scanned Legislative Manuals, and I haven't read their totals, so I won't sign my name to that line yet.

One older transition is on the record. The 1970 canvass lists Warren Spannaus as the DFL candidate and Robert A. Forsythe as the Republican, and has Spannaus winning 680,479 to 644,390, about 2.72 points of the two-man vote. The Legislative Reference Library's roster has Spannaus taking office January 4, 1971, succeeding Douglas Head, who took office January 2, 1967. Since then the office has had five occupants: Spannaus, Hubert H. Humphrey III, Mike Hatch, Swanson and Ellison.

The best case for each reading

If you're with Ellison, here's your best case. Two pollsters using different methods landed on the same 8-point lead in the same week. State Navigate weighted to party, region and who'd already voted, among other things, and still got 52 to 44. SurveyUSA, which listed no party target in the releases that printed its weighting, got 49 to 41. SurveyUSA's 2022 miss was a miss in the Republican's favor, and the whole miss sat on the incumbent's side. And Schutz's ceiling in seven SurveyUSA polls was 45.

If you're with Schutz, here's yours. The final SurveyUSA poll's Democratic edge doubled in two weeks, and putting September's party mix back cuts the lead roughly in half. Morning Consult had it at 4 among registered voters in early October, with 14 percent undecided. In 2022 Ellison ran about 7 points behind the top of the ticket on the real ballots. Each pollster's latest 2026 poll has him only 2 to 5 behind Klobuchar, who led by 9 to 12 in those same polls. If the 2022 gap holds, that's a 2-to-5-point race, not an 8. And nearly one in five independents in the final SurveyUSA poll hadn't decided.

Both cases come out of the same releases. Neither one's making anything up. The data don't pick a winner. Voters do.

What the numbers can't carry

Down-ballot races pull a lot of undecided voters. That's about the race, not a knock on any one pollster. In the final SurveyUSA poll, 10 percent were undecided on attorney general, against 8 percent for governor. In July it was 20 percent. SurveyUSA said of the Secretary of State's race in September that "the relatively lower profile of this race means any outcome will be possible." The same goes here, with half the undecideds.

And I'm the last guy who gets to be smug about numbers. I got 15 percent in August.

Then the count

Unofficial election-night count, as of 12:57 a.m. Central on November 4, 2026, with 3,986 of 4,105 precincts reporting.

Here's the count, or the 97.1 percent of precincts that had reported. The Secretary of State's election-night file has Ellison at 1,503,996 votes, 54.44 percent, and Schutz at 1,254,573, 45.41 percent, with 4,189 write-ins. That's a lead of about 9.0 points of the two-candidate vote, with 119 precincts still out. It isn't the result yet. It's most of it.

Now put it next to the table up top. SurveyUSA's final poll had Ellison up 8. So did State Navigate. Morning Consult, about four weeks out and among registered voters, had him up 4. The two late polls came in about a point under the count. Four years ago SurveyUSA's final missed by 7.84 points in the same direction. I'll take a one-point miss over an eight-point miss every day of the week.

Same caveat as before, because it still matters: the polls' numbers include their undecideds and the count doesn't. SurveyUSA had Ellison at 49 and Schutz at 41. On the count they're at 54.44 and 45.41. Ellison came in 5.44 points above his poll number and Schutz 4.41 above his. In 2022 the whole miss sat on the incumbent's side. This time it's on both sides, a little more on Ellison's.

Remember the gap between the two races. In 2022 SurveyUSA's final poll had Ellison 15 points behind Walz, and on the ballots he ran about 7 behind. This time its final poll had him 4 behind Klobuchar. On the count, Klobuchar's two-candidate lead over Demuth is 13.37 points and Ellison's is 9.04, so he's running about 4.3 behind. The poll got the gap about right. That also answers the best case I laid out for Schutz's side up top. On tonight's count, the 2022 gap didn't hold.

Where the eight points of swing came from

Statewide, the margin moved 8.21 points toward Ellison from the certified 2022 race. It moved toward him in every region, and it moved most in the suburbs.

Ellison lost those five suburban counties combined in 2022. He's carrying them by about 6 tonight. Outside the metro he's still losing by almost 18, but that's five points better than four years ago, and five points across 80 counties is a lot of kitchen tables.

Thirteen counties hadn't finished reporting at this snapshot, so the county figures below are limited to counties that had, plus one that was nearly there. Four counties Schultz carried in 2022 are going to Ellison: Carlton (−2.8 to +2.4), Rice (−2.0 to +2.2) and Winona (−4.1 to +3.4), all fully reported, and Anoka (−10.0 to +1.4), with 126 of its 128 precincts in. Anoka is the one that jumps off the page. It's the biggest of the suburban counties Ellison lost last time, and it went from a 10-point loss to a narrow lead.

The biggest swings among fully reported counties were Nobles (13.41 points), Mahnomen (12.88), Clay (11.29) and Dakota (10.45). Anoka, nearly complete, swung 11.39. Not all of those are wins. Nobles still went Schutz by 24.23 and Mahnomen by 5.57. A swing is movement, not a flip.

The two races moved together

By county, Ellison's margin and Klobuchar's margin line up almost perfectly: a correlation of 0.999, against 0.995 for Ellison and Walz in 2022. The swings line up too, at 0.859. Where the top of the ticket gained, the attorney general's race almost always gained with it.

The gap between them shrank. Across counties, Ellison ran a mean 3.11 points behind Klobuchar, unweighted. In 2022 he ran 4.22 behind Walz. That's still a gap, and it's a smaller one. These comparisons run across all 87 counties, the 13 partial ones included, so they can shift a little as the rest comes in.

What can still move

This is the unofficial election-night count. Ballots counted after election night show up only as counties report them. The counties with precincts still out at 12:57 a.m. were Cass, St. Louis, Olmsted, Faribault, Swift, Isanti, Washington, Anoka, Morrison, Roseau, Carver, Kanabec and Sherburne, so their numbers can move, and so can the regional totals that include them. Under Minn. Stat. § 204C.33, the county canvassing boards meet "between the third and eighth days following the state general election", and the State Canvassing Board meets "on the 16th day following the state general election", which this year is Thursday, November 19, unless that's a state holiday. When the board certifies, this piece gets the certified margin, and I'll grade the polls against that.

What we can do

None of this needs a law. It needs the people who pay for polls, and the people who print them, to show their work.

Print the party mix next to the topline. SurveyUSA already prints it at the bottom of every table. KSTP put it in its final story. Make that the rule for every outlet that reports a Minnesota poll, not a nice surprise. A reader who sees "Ellison +8" should see "in a sample that's D+8" in the same sentence.

Say whether the poll was weighted to party. SurveyUSA's own methods statement says weighting to party ID depends "on client preference." State Navigate says it weighted to leaned party. That choice can move a race by several points. If it's the client's call, the client should tell us what it called.

Report the error on the lead, not just on each number. A ±4.2 interval on each candidate is a much wider band on the gap between them. Put that in the story, right next to the lead.

Grade the final poll against the certified count, every cycle, in public. The State Canvassing Board declares the result within three days after it finishes its canvass. The pollster's final release is a PDF. Put the two side by side and publish the miss. I'll do it for this race here, as soon as the board certifies.

A poll is a picture of whoever answered. The count is everybody who showed up. Tonight we got most of it.

First the facts. Then the fix.

Sources

The SurveyUSA releases were read in full as the pollster's own crosstab PDFs, which KSTP hosts: Election Poll #27951 (July 15–20, 2026, at kstp.com/wp-content/uploads/2026/07/ElectionPoll.pdf, which also reports the June result), #27957 (July 29–Aug. 4, .../2026/08/ElectionAugust.pdf), #27966 (Aug. 13–17, .../2026/08/Election-Poll-8-19-26.pdf), #28000 (Sept. 9–14, .../2026/09/Election-Poll-9-16.pdf), #28052 (Oct. 8–12, .../2026/10/Poll-Races-1.pdf) and #28092 (Oct. 24–28, released Oct. 29, .../2026/10/PollMargins.pdf). The party-mix and subgroup figures are from the "Composition of Likely November Voters" rows and subgroup columns of each release's Attorney General table. KSTP's October 30, 2026 story on the final poll is the source of the sentence quoted from KSTP. The 2022 SurveyUSA releases are #26468 (Aug. 30–Sept. 4, 2022, .../2022/09/SurveyUSA-Poll-governor-statewide-races-09-07-22.pdf) and #26632 (Oct. 26–30, released Oct. 31, 2022, .../2022/11/KSTP-SurveyUSA-election-poll-11-1-22.pdf). SurveyUSA's methods statement is quoted from the statement printed in those releases.

State Navigate's Minnesota statewide poll of October 26–28, 2026 was read as the pollster's PDF at projects.statenavigate.com. The Morning Consult figures are from The Daily Agenda's October 9, 2026 newsletter, which sponsored the poll and published its results.

The 2022 results are from the State Canvassing Board Certificate for the 2022 state general election (filed November 29, 2022), read through the Internet Archive's copy of officialdocuments.sos.mn.gov, document 133584; its 87 county rows sum to its statewide totals. The 1956, 1958, 1970, 2014 and 2018 results, including the 1970 totals on page 2 of the Library's PDF of the 1971–1972 Legislative Manual, are from the official canvass tables in the Minnesota Legislative Manuals posted by the Legislative Reference Library at lrl.mn.gov. The office's succession dates are from the Library's roster, "Minnesota Territorial and State Attorneys General, 1849-present". The canvass timetable is Minn. Stat. § 204C.33, read at revisor.mn.gov on November 3, 2026. The decomposition of the September-to-October swing, the subsample estimate, the 2022 poll misses and the two-candidate margins are my arithmetic on those records. The 2026 results are from the Secretary of State's election-night result files at electionresultsfiles.sos.mn.gov (the statewide Attorney General and Governor files and the all-races-by-county file), read at 12:57 a.m. Central on November 4, 2026, with 3,986 of 4,105 precincts reporting. They are unofficial. My August 11, 2026 primary result, 15.33 percent, is from the Secretary of State's post-canvass results at electionresults.sos.mn.gov. The Secretary of State's "About These Results" page, read the same night, says counties that receive absentee ballots after 3:00 p.m. on election day may choose to delay processing and reporting them, that the county must update its results after processing them, and that the State Canvassing Board meets Thursday, November 19, 2026.

What I searched, what I found, and what I could not establish. An October Star Tribune/KARE 11 Minnesota Poll by Mason-Dixon also surveyed this race; the paper's page refused my requests and I found no copy of the pollster's release, so that poll isn't in the table. Morning Consult's own release and crosstabs weren't available to me; the sponsor's linked crosstab image sits behind a sign-in, so the Morning Consult row and subgroups rest on the sponsor's published text. The sample size and interval for SurveyUSA's June poll aren't in any release I read. I didn't use race ratings from forecasting sites; the one I tried to open blocked the request. Of the thirteen attorney general elections from 1962 through 2010, I read the statewide total only for 1970; the other twelve are in the Library's scanned manuals, and I didn't read their totals, so whether 2022 was the closest race since 1956 is open; 1954 also wasn't read. The 1956 and 1958 manual lines don't print party labels. KSTP's report of a 41 percent approval rating for Ellison isn't in the pollster's release, so it isn't used. SurveyUSA's releases don't define the counties in each region, so its regional numbers aren't compared with county results. The 2026 figures are the unofficial election-night count at 12:57 a.m. Central on November 4, 2026, with 3,986 of 4,105 precincts reporting; 119 precincts in 13 counties (Cass, St. Louis, Olmsted, Faribault, Swift, Isanti, Washington, Anoka, Morrison, Roseau, Carver, Kanabec and Sherburne) were still out, so named-county figures are given only for fully reported counties and Anoka, which had 126 of 128 precincts in; the regional totals and the county correlations include all 87. The regional and county margins, the swings, the flips and the correlations were computed with a script that recomputes each county's two-candidate margin from the 2022 certificate and the 2026 county file; I don't report its turnout comparison, because a partial count isn't comparable with a certified one. Corrections: campaign@madgettformn.com.