Pollsters Were Wrong About The Midwest - Here’s Why

Polls suggested Abdul El-Sayed was cruising toward the Democratic Senate nomination in Michigan. In Wisconsin, surveys repeatedly gave gubernatorial candidate Francesca Hong enormous leads.

Election night produced something considerably different.

El-Sayed did win Michigan's Democratic Senate primary, but by just one percentage point, defeating Rep. Haley Stevens 48.5% to 47.5%. Hong did not win at all. David Crowley defeated her by less than one percentage point in Wisconsin's Democratic gubernatorial primary.

Those results left pollsters examining how surveys could miss two major races by such substantial margins. Their explanations point to several problems particular to primary polling: open primaries, incorrectly modeled electorates, late-deciding voters and an apparent tendency for younger, more ideological voters to be more willing to answer surveys.

“The polls in Wisconsin and Michigan did miss in a big way,” Eli McKown-Dawson, senior elections analyst at Silver Bulletin, told the Daily Caller News Foundation.

But primary polling, he cautioned, is inherently volatile.

“The average polling error in polls of presidential primaries historically is about 9 points,” McKown-Dawson said. “It’s even higher, easily in the double digits on average, in polls of House and Senate primaries.”

Michigan and Wisconsin present another complication because both have open primaries. Voters do not register by political party, making it considerably harder for pollsters to determine beforehand who will actually participate in a Democratic primary.

That matters because modeling the electorate is fundamental to polling. A survey can accurately record what its respondents say while still producing a misleading result if those respondents do not resemble the people who ultimately vote.

That appears to have been a significant problem in these races.

Before Michigan voted, surveys from Emerson College Polling/WLNS/WOODTV and Mitchell Research & Communications showed El-Sayed leading Stevens by more than 20 points.

Instead, he survived by one.

Democratic pollster Adam Carlson, founding partner of Zenith Research, wrote in a post-election analysis that El-Sayed's supporters appeared especially eager to participate in surveys.

“El-Sayed’s supporters were more fired up and willing to take polls down the stretch, but more enthusiasm does not necessarily equal more votes,” Carlson wrote.

Carlson identified several other shortcomings in his Michigan polling, including inadequate representation of seniors and voters earning less than $50,000 annually. He also raised the possibility that some Republicans participated in the Democratic primary and supported Stevens to prevent El-Sayed from becoming the nominee.

The actual results revealed an electorate different from the one some polling had anticipated.

Stevens performed better with older, non-college-educated and lower-to-middle-income voters. El-Sayed's strongest results came in counties with larger white, college-educated and higher-income populations.

The ideological contrast between the candidates was pronounced.

Stevens campaigned as the more moderate Democrat, supporting reform of Immigration and Customs Enforcement rather than its abolition, favoring a public health insurance option instead of Medicare for All, and maintaining strong support for Israel.

El-Sayed received backing from Sen. Bernie Sanders, Rep. Alexandria Ocasio-Cortez and Michigan Rep. Rashida Tlaib. Senate Minority Leader Chuck Schumer quietly backed Stevens and encouraged donors to support her, The New York Times reported.

Wisconsin's polling miss was even more striking.

Surveys from State Navigate, Marquette University and Public Policy Polling had repeatedly shown Hong leading Crowley by more than 20 points. Crowley ultimately won by less than one point.

One of the clearest problems was age.

Pollster G. Elliott Morris found that, among Wisconsin primary voters whose ages were known, approximately 52% to 53% were seniors. Just 2% to 3% were younger than 30.

Compare that with State Navigate's July polling sample: 36% seniors and 17% under 30.

The survey electorate was dramatically younger than the electorate that actually showed up.

“It looks like both of these processes failed to anticipate an older, more moderate primary electorate,” Morris wrote.

That imbalance was particularly consequential because younger and more enthusiastic respondents were disproportionately supportive of candidates such as Hong and El-Sayed.

The Wisconsin race also illustrates another problem that no statistical weighting system can entirely eliminate: voters can change their minds.

A Marquette Law School poll found 34% of respondents were still undecided in late July. Crowley apparently benefited heavily as those voters made their decisions.

There had also been considerable movement among the candidates. Candidates dropped out late, and Crowley reentered the contest only weeks before Election Day.

A poll conducted before that movement could accurately describe voter preferences at the time it was taken and still be badly wrong about the final result.

Morris therefore cautioned against treating primary surveys as predictions.

“The right reading of this is that primary polling is hard in ways that general-election polling is not,” he wrote, describing primary polls as “a snapshot with a very short shelf life.”

Hong also faced scrutiny late in the campaign over previous social-media posts.

A 2020 post in which she called for Thanksgiving to be canceled and urged Americans to stop “celebrating colonialism” resurfaced during the race. On Election Day, Hong said she would not cancel Thanksgiving.

Former Obama adviser and “Pod Save America” co-host Dan Pfeiffer argued that electability became increasingly important during the campaign's closing stretch. He pointed to Crowley and Wisconsin Gov. Tony Evers emphasizing the issue, while Hong faced questions about her previous posts.

Sanders and Ocasio-Cortez, despite their support for other progressive candidates, did not campaign with Hong before the primary.

None of this means polls are useless, nor does it establish that progressive candidates inherently poll better than they perform. The evidence from these two contests points to more specific problems.

Pollsters surveyed electorates that were too young. Highly enthusiastic supporters were particularly willing to respond. Open-primary rules made predicting participation difficult. And in Wisconsin, a large bloc of undecided voters apparently moved late.

The distinction is important heading into November.